What is best projection for network analysis

[MDKaller]

I can use some advice.  I work for a school district, where I must calculate the walking distance of a school using Network Analysis by creating a service area using distance as a cost.  But I am always try to make the most accurate findings.  Before, I used either UTM NAD 83 12N or the WGS 84 Web Mercator.  But I discovered that USA Contiguous Conformal Conic preserves true distance, so I've used that in projecting my roads layer.

And yet, I get calls from parents and others who use Google maps or what have you to calculate the distance.  And I am aware that there are discrepancies between the different projections.  So I have some questions:

1.) What geographic projection or coordinate system is ideal for roads layer in terms of distance as a cost for network analysis?

2.) How to reconcile the discrepancies between what ArcGIS PRo versus online mapping like Google maps?

[TravisSaladino]

Yeah taking a mostly spherical object (the earth) and making it flat (a map) can cause all kinds of issues... or distortions in shape, area, distance and/or direction. In most consumer types maps the projection is WGS 84 Web Mercator. The Mercator is the important part there because it distorts distance and area to preserve direction. Back in 1569 you needed to make sure your boat was going in the right direction more so than anything else. So now when you use that projection and measure distance (and therefore area too) further away from the equator the larger the distortion of distance on the map. That's why the parents think their kid has a .3 mile walk and you tell them it's actually .25 miles. 

And by the way, a geographic coordinate system isn't appropriate for any measurements either because the units used are not fixed like feet or meters are. Geographic coordinate systems use angular units like decimal degrees or Degree Minutes Seconds, the distance of which on the earth change depending how far you are from the equator too.  

So that really answers your second question. In ArcGIS Pro you are leveraging a projection designed to minimize the distortion of distance at the cost of distorting direction or shape. In most web maps (Google set this expectation of look by the way with their first consumer maps) are used around the world and the considerations for distortions are different for that scale of data and scope of audience.

Now let's tackle your first question. I would ask you, what is the extent of the network you need to model? If it's "local" use a coordinate system designed to minimize all distortions over a local area like a state plane system in the US, or UTM or some other local system outside the US. If the extent is larger, like all of North America, use something like USA Contiguous Conformal Conic. 

Lastly, remember you are modeling the earth in GIS not reconstructing it 1:1. There will be errors, it's just what error can you live with. 

Hope that helps.

[AlanHatakeyama]

Hi MDKaller,

Generally speaking, I would create my cost attribute based on a field that I calculated to be the geodesic length of each street segment.  This way, the actual geodesic length of each street segment is used, instead of having it based on the planar length of the projected coordinate system that may be subject to distortion.

The downside of doing this is that if you modify the geometry of any of your street features, you need to remember to recalculate the length field so that it contains the updated length of the street.  Likewise if you create new street features, these new streets need to have their length field calculated.

If you are working within a relatively narrow area (i.e., entirely within a few degrees of longitude), the UTM projections work great for a single projected coordinate system.  In your example, the 12N zone covers longitudes between 108°W and 114°W in the northern hemisphere.  UTM zones have a very small error (about 0.1%) for features within its defined narrow range of longitudes.

As TravisSaladino alluded to, I would avoid using the Mercator projection unless you are within a couple of degrees of latitude from the equator; as the distortion significantly increases the further away you get from the equator (at 33.5°N latitude (near Phoenix, AZ), 1000 meters in the real world is modeled as about 1200 meters in the Mercator projection; and at 47.5°N latitude (near Great Falls, MT), 1000 meters in the real world is modeled as about 1500 meters in the Mercator projection!).

As for the USA Contiguous Lambert Conformal Conic projection, I would somewhat hesitate due to it covering such a large range of latitudes needing to cover much of the continental United States.  This projection has its Standard Latitudes defined at 33°N and 45°N, meaning that measurements are exact at these latitudes but degrade as you drift away from them.  At the Central Latitude of 39°N (halfway between the Standard Latitudes), the error is about 1%, which is good but not great (it depends on how accurate you really need to be).  Down in southern Florida at 25°N latitude, the error is a whopping 2.5% or so due to that area being so far removed from the "main land mass" of the continental United States as a whole.

As for reconciling discrepancies between your analysis and mapping system found online, you may want to try comparing a ground-truth measurement to both your analysis and the other online mapping system and see how each compares to the real world measurement.  Your street network may have a road or walkway that doesn't exist in the other online mapping system (or vice versa) allowing a student to "cut through" in one solution but not in the other.  It may be the case that connectivity between intersecting streets is not properly modeled in one system vs the other, causing the routes to differ in length.  Just some other things to consider besides the measurement differences due to projections.

Please let me know if you have any questions.  Thanks!

Alan