Hi Tim,

 

Thanks for contacting us about this.

 

We have encountered this scenario twice in the last six months. One case was your exact scenario: buffer animal locations, sample oceanography within the buffers, and obtain mean values. The other involved sampling the region covered by drifting longline fishing gear. Although MGET helps with this scenario, we haven't yet written a tool that handles it directly.

 

When my colleague was working on it, I suggested the following strategy for sampling a given time series of rasters:

 

1.    Using Spatial Analyst Create Constant Raster, create a raster that has the extent and cell size of the rasters. (Each raster in the time series must have the same extent and cell size.) For optimal performance, you may need to clip it down to your study area.

2.    Use Raster to Point to convert that raster to a grid of points representing the centers of the raster cells.

3.    For each telemetry location:

a.    Buffer it

b.    Intersect the buffer with the grid of points, to get the points that are in the buffer

c.    Sample those points using MGET or Spatial Analysts's Extract Values to Points tool

d.    Use the Summary Statistics tool to get the mean value of all the points

 

My guess is that most people who attempt this must resort to programming.  Steps 1 and 2 you can obviously do without programming. It may be possible to implement step 3 without programming and without doing a loop. For example, step a can be achieved by buffering all the points using Dissolve Type of NONE to produce overlapping polygons. The crux of doing all of step 3 without a loop is step b: I don't remember what happens if you intersect points with overlapping polygons: are two points created? If yes, then you're good to go. If not, then, ultimately, you must call intersect once for each telemetry location.

 

In 3c, if you implement without doing a loop, then you'd want to use MGET's Sample Rasters Listed in Fields tool, similar to the example on the MGET web site.

 

The performance of this will be pretty slow. My colleague ended up sampling 600,000 points in step 3c and it took many hours. I am planning to rewrite MGET's sampling tools in the future to speed this up as much as I can, probably by not even calling ArcGIS to perform the sampling and using something like GDAL to read the rasters directly.

 

This approach suffers if the buffers do not enclose many grid points. In that case, you may want to know how much of a given cell is enclosed, and do your averaging based on that. This would require an even more complicated solution.

 

If this approach sounds like what you want, I'm happy to discuss more details with you.

 

Best regards,

 

Jason