Best spatial references for models using distance as primary input

[Jay_Gregory]

I am working with a model that is using distances between points as a significant input.  My initial study area is in Texas, but eventually will cover portions of the entire US (mostly urban areas).  

My question is: what should my projections be for my two point feature classes?  Currently I'm using WGS 1985 WebMerAuxSph (3857) as my projected coordinate system and WGS 1984 (4326) as my geographic coordinate system.  

When I am generating near tables I am using planar distance, but could also use geodesic.  

The distances in question for each study area shouldn't be more than a 5 or 10 miles between points, but could be as close as 10m (or 0 probably).  

Any suggestions on the most accurate projections to use?

[Sean_Wray]

I don't know what the best would be, but it seems to me that if you are concerned about measurement accuracy you could divide you data into sections and use more localized coordinate systems such as a state plane coordinate system or a UTM system. With UTM there are 10 sections that cover the  entire US.

In central Texas I know the difference between measurements between a Web Mercator measurement and State Plane measurement that runs east/west can up to 7 or 8 feet. 

I'll be interested to see what others who are smarter then me think.

[AndyAnderson · Accepted solution]

I agree with Sean — don’t use planar distances with Web Mercator. It will be even less accurate the farther north you go. Geodesic distance will be the most accurate but it will be slower to calculate. But that still might be faster than recasting two points to UTM (good and easy to choose the correct projection) or state plane (better but harder to choose the correct projection) followed by a planar distance calculation.

[DanPatterson]

When distances are less than 3 degrees apart, I tend to use UTM ( lune width 6 degrees, hence is 1/2 a lune width, scale factor 0.9996 at the central meridian).

If closer than that, I use Modified Transverse Mercator (state plane in the US, I think) since it uses a 3 degree lune and the scale factor is 0.9999 at the central meridian.

As for "geodesic" calculations, I am not sure what esri uses since I can't find specific documentation on it (eg are the calculations  Haversine-like, Vincenty-like or some other variant that exists out there).

More importantly, what precision and accuracy do you require for your distance calculations?  and how accurately and precisely are your locations recorded?  Answer those 2 and you can always find a projected coordinate system or a geodesic calculation method that will meet your needs.

[Jay_Gregory]

Thanks all - this is valuable input. I need to talk with the folks that developed the model but this is extremely helpful. 

[Jay_Gregory]

@DanPatterson @AndyAnderson @Sean_Wray 

Follow up question: if I am using high resolution terrain data (SRTM) to extract elevation values for my points as well, would it follow I should be projecting the individual tiffs to the same UTM or StatePlane projection I am using for my point feature classes before extracting elevation from the rasters?