2

I am using ArcGIS 10.1 for Desktop.

I have a value raster and a polygon feature in GCS coordinates. My raster resolution is 0.5 decimal degrees which is much larger than the size of my polygons. My value raster has discrete floating points. So I resampled my raster to smaller resolution of 0.01 using the NEAREST option and used Zonal Statistics in table to get the mean value for each polygon. However when I compare the means of the bigger polygons that have zonal statistics values for both original and resampled raster, I find that the resampled values are higher. My raster values are in terms of mm/sq km. So the small difference during resampling is amplified when I scale the value up with the polygon areas. Any ideas how to prevent this?

I have also tried doing the same with projecting both my raster and polygons to WGS and then resampling and zonal statistics. But i get the same trends in the results. Am I missing something?

Following are 2 graphs showing: 1) comparison of regional zonal means evaluated using original GCS raster and evaluated using resampled/projected+resampled rasters 2) comparison of country level values (zonal mean*area) and sum of all district values

It seems that the country level and sum of district level values match each other. But both are generally higher for resampled and projected rasters. So resampling is giving me higher zonal means.

Comparison of zonal means using original raster (GCS 0.5 deg resolution) with zonal means using resampled/projected+resampled rasters. The numbers in the labels indicate cell size used for resampling Comparison of country level and sum of district level values for (zonal mean * polygon area)

  • Certainly the new means will change, because the resampling usually changes the weights of cells that are only partially covered by any polygon. (Isn't that the entire point of the procedure in this instance?) As a consistency check, you could multiply the polygon areas by the zonal means and total those over the polygons. These totals can change, too, but they should not change a whole lot--and if you include a surrounding "universal" polygon in the calculation, the only differences should be due only to floating point (single precision!) error. Do you observe that consistency or not? – whuber Dec 1 '15 at 0:41
  • When I multiply the zonal means by the polygon areas, my values are almost twice that of what I expected. I did a total at the country level using a polygon with just the country boundary and compared that to the total I get for the regions within the country. I'm interested in only 75% of the regions within the country. I'll try doing the calculation again using the Union of both regions and country boundary as you suggest with the universal polygon. Maybe that will help – Cicatrixx Dec 1 '15 at 7:11
  • @whuber I have added some figures showing my comparison. I still see the zonal means are higher with resampled raster. Just wondering, my raster values are between 0-1. Does that make a difference perhaps? – Cicatrixx Dec 1 '15 at 9:48
  • Involving projections is complicating the issue, because now we have to wonder about the interactions among the reprojection (which inherently resamples), the resampling, how areas are computed, and exactly how you are carrying out all these operations. It would be a good idea to isolate the problem by sticking to one coordinate system and studying how just the resampling affects the results. – whuber Dec 1 '15 at 14:56
1

So I tried resampling my value raster with even finer resolutions, both in GCS and projected version. I ran zonal statistics on each resampled raster to get mean for my regions. Then I summed the mean*polygon area for all my regions. The attached graph shows the comparison of the sums. And I can see that my resampling cell size earlier wasnt fine enough. After making the cellsize super fine the total sum doesnt change anymore. SO that solves my problem I guess.

What do you think @whuber?

Comparison of total sum for different cell sizes

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.