Terrain aspect similarity

Compute the difference, and then take 360-the absolute difference if the absolute difference is greater than 180.

Here's one way of doing that in an R function:

d180 = function(a,b){
 d = a-b
 d = ifelse(abs(d)<180 , d , 360-abs(d))
 d
}

and some examples:

> d180(0,0)
[1] 0
> d180(0,1)
[1] -1
> d180(0,180)
[1] 180
> d180(0,181)
[1] 179
> d180(358,2)
[1] 4
> d180(2,358)
[1] 4

Whether the distance is positive or negative is a bit odd, return the abs(d) if you don't care...

A bit of a different perspective, you could just linearize aspect. Here is pseudo-code pulled from the linear aspect tool in our ArcGIS Geomorphometry & Gradient Metrics toolbox. Note; we use (mod (..) * 100) and (36000(360*100) / 100) allow for two decimal points since Fmod is no longer available in ArcGIS.

a = aspect(elev) # degrees
t1 = focal(sin(a/57.296), window=3,"sum")
t2 = focal(cos(a/57.296), window=3,"sum")
linear.aspect = mod(((450-(atan2(t1, t2) * 57.296)) * 100), 36000) / 100

Another nice trick is to look at the Stage (1978) slope-aspect transformation that provides an interaction. Assuming slope in degrees (Stage paper assumes percent) and aspect in degrees, you can calculate it thus:

s = (slope(elev) / 0.572957795786) * 0.01
a = aspect(elev) * (pi / 180)
scosa = s * cos(a)
scosa = s * sin(a)

Example transformed values for 50% slope across 10 aspects.

Aspect    cosine    sine
N         0.500     0.000
N30E      0.433     0.250
N45E      0.345     0.345
N60E      0.250     0.433
E         0.000     0.500
ESE      -0.354     0.354
S        -0.500     0.000
SSW      -0.354    -0.354
W         0.000    -0.500

Or, linearize with a recentered origin. This is done in vegetation modeling to provide a metric that is optimized for site prodictivity (eg. Roberts and Cooper 1989 TRASP transformation) Assuming degrees, 1 - cos( (pi / 180) (a - 30)) / 2

a = aspect(elev)
trasp = (1 - cos( (3.142 / 180) * (a - 30)) ) / 2 

These transformations are available in ArcGIS in the Geomorphometry & Gradient Metrics toolbox and in the R package spatialEco (on CRAN).

Retaining sign (pseudo):

a = modulus((targetangle-sourceangle),6.283185307179586)
if (a>3.141592653589793)
    return a-6.283185307179586
else 
    return a

(Talking degrees:) If modulus(-45, 360) doesn't return 315 in your language, then use modulus((-45 + 360), 360). Your modulus fn must support decimal values.