Project Rasters- What are the best practices for continuous data?

[JoshuaBrengel]

I'm trying to project a PRISM precipitation raster to match a vector data's projected coordinate system so that I can perform zonal statistics on it.   The original raster data has a GCS of NAD 1983 and I'm trying to project it into Albers Equal Area Conic.  What should be the anticipated effect of using the Project Raster GP tool (ArcMap 10.4, SP1) on my original raster dataset?  

It appears that the cell size, extent, and spatial location have all changed between my original dataset and my projected dataset.  I used measure tool to check cell size.  Also, the suggested resampling technique for continuous precipitation data should either be cubic of bilinear, but regardless of what resampling I choose, there are changes to the original dataset.

I'm not sure if these changes are to be expected and acceptable.  From my couple days of research and discussions with coworkers, there are not a lot of available resources on the topic of raster projection and its best practices. 

Should I just scrap the idea of projecting the raster and use the dataset's original GCS for my zonal statistic analysis?

[DanPatterson_Retired]

What was your cell size in decimal degrees? and did you approximate its equivalent in meters?

Decimal degrees - Wikipedia the cos(lat) modifier

a cell surrounded by cells of the same value should remain the same, you may get differences at the edge of zones, but how accurate is the original raster representation in the first place?

If you want, you could just use zonal statistics as table and work from the counts, converting to unit area after.

[JoshuaBrengel]

Thank you very much for the quick reply Dan.  The original raster's cell size in DD is 0.04166667 x 0.041666667. Projected raster (to Equal Albers) is 4676.765775.   Just using the 1 degree = 111,139 m w/out modifier, that is pretty close to each other. 

I'm assuming the concept of a modifier becomes important when I use the "Measure Tool" to get a geodesic line measurement.   That's why even if the cell size says 0.0416 DD x 0.0416 DD in properties of original unprojected raster, when I measure the cell size of cell located in Florida in original raster it is measuring wider than a cell located in NY.

Are there any other best practices I should consider when using the Project Raster tool?  For example, should I set a snap raster or set my extent to the original raster in envt setting?  I'm already assuming pick proper resampling technique for type of data (continuous vs discrete).

Finally, any suggestions on resources for practical considerations for projecting rasters so that I have baseline expectations for should happen to my raster data upon projection?  

I realize this a few questions, though I think others could really benefit from your considerations!  Thanks!

[DanPatterson_Retired · Accepted solution]

The modifier is for the E/W (aka latitude) factor since lines of latitude (aka, parallels) have to get smaller as you move poleward. 

How accurate/representative/precise are your input values in the first place?  (pick any of those terms, I doubt that your input data are 'continuous' to begin with but measured at discrete locations and interpolated.  aka... precipitation, is never a 'continous' surface, it is an 'interpolated' surface but represented as continuous-ish.

As for the need to project in the first place, you indicate that you are needing 'zonal statistics'

So I would be as concerned about the geometry representation of your zones.  Projected or unprojected, your 'zones' are only as good as the geometry used to make them.

I think "cell center" is the determining factor as to whether a cell is 'in' or 'out' of a zone, so the bigger the cell, the greater the potential for over or under representation … the latter is hard to get a handle on unless you experiment with zones of various shapes and configurations.  For a example a long skinny zone may not get 'hit' by an overlain raster with a large cell size.

You could experiment... halve your cell size, increases the spatial resolution by 4X... not your accuracy, but the resolution, the hit/miss of a cell center and whether a cell will fall within one zone or another.

Lots to consider, but nothing will make a more "accurate map/analysis" just one that is "maybe ess wrong".  Just qualify your conditions and draw your discussion within those bounds of qualification.

[JoshuaBrengel]

Thank you Dan, this is helpful.  I believe the PRISM dataset is, generally speaking, based on station data that's been interpolated. 

Your zonal stat suggestions are appreciated.  It certainly feels a bit overwhelming when considering all the variables that can change in not only projecting a raster, but also then further analyzing it via zonal statistics.  So it's good to know that shooting for "less wrong" is an appropriate way to think.