Take the 2-minute tour ×
Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It's 100% free, no registration required.

I am wondering if anyone can clarify whether point data should be projected to an equidistant projection when using ArcGIS IDW spatial interpolation?

I am working on a dataset from Western North America spanning about 30 degrees of latitude. The data are currently in Lat/Long (NAD83).

Does ArcGIS "project on the fly" or somehow adjust for latitude when calculating distances to my sample points during the interpolation procedure or should I be supplying everything in a projection that preserves distances?

share|improve this question
2  
No projection preserves all distances. An "equidistant" projection preserves distances to one, two, or at most three distinguished points on the map. In order to do this, it typically introduces a large distortion in distances among other pairs of points. See this answer and the comments to it. –  whuber Apr 16 '13 at 2:40

2 Answers 2

up vote 8 down vote accepted

IDW works by finding the data points located nearest each point of interpolation, weighting the data values according to a given power p of the distances to those points, and forming the weighted average. (Often p = -2.)

Suppose there is some amount of distance distortion around an interpolation point that is the same in all directions. This will multiply all distances by some constant value x. The weights therefore all get multiplied by x^p. Because this does not change the relative weights, the weighted average is the same as before.

When the distance distortion changes with direction, this invariance no longer holds: data points in some directions now appear (on the map) relatively closer than they should while other points appear relatively further. This changes the weights and therefore affects the IDW predictions.

Consequently, for IDW interpolation we would want to use a projection that creates roughly equal distortions in all directions from each point on the map. Such a projection is known as conformal. Conformal projections include those based on the Mercator (including Transverse Mercator (TM)), Lambert Conic, and even Stereographic.

It is important to realize that conformality is a "local" property. This means that the distance distortion is constant across all bearings only within small neighborhoods of each point. For larger neighborhoods involving greater distances, all bets are off (in general). A common--and extreme--example is the Mercator projection, which is conformal everywhere (except at the poles, where it is not defined). Its distance distortion becomes infinite at sufficiently large north-south distances from the Equator, while along the Equator itself it's perfectly accurate.

The amount of distortion in some projections can change so rapidly from point to point that even conformality will not save us when the nearest neighbors are far from each other or near the extremes of the projection's domain. It is wise, then, to choose a conformal projection adapted to the study region: this means the study region is included within an area where its distortion is the smallest. Examples include the Mercator near the Equator, TM along north-south lines, and Stereographic near either pole. In the conterminous US, the Lambert Conformal Conic is often a good default choice when the reference latitudes are placed within the study region but near its northern and southern extremes.

These considerations usually are important only for study regions that extend across large countries or more. Within small countries or states of the US, popular conventional coordinate systems exist (such as various national grids and State Plane coordinates) which introduce little distance distortion within those particular countries or states. They are good default choices for most analytical work.

share|improve this answer
1  
Thank you for the detailed response. I've rerun the analysis after projecting to LCC. There are slight differences from the unprojected data so I'll go with this new projected version. –  Julie Apr 16 '13 at 20:25

Since you are using an operation that deals with calculating distance your input dataset should be in a projected coordinate system. I would suggest re-projecting your point data to a UTM projection (see UTM reference map).

share|improve this answer
1  
UTM is appropriate when the range of longitudes is small: officially, less than seven degrees. (At more extreme latitudes you can get away with a bit more range.) The Western US is large enough that a UTM coordinate system is not likely to be the best choice or even a good one. –  whuber Apr 16 '13 at 2:43
    
I suggested UTM because the question talked about the data having a greater range of latitude than longitude. –  artwork21 Apr 16 '13 at 12:04

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.