[GRASSLIST:9042] Re: distance algorithim

[Joel Peter William Pitt]

On 11/9/05, Hamish <hamish_nospam at yahoo.com> wrote:
> how do the results compare between the two methods? i.e. how good a job
> does r.cost do with the knight's move? without?
>
> r.mapcalc residuals=map1-map2
> r.univar residuals
>
> sort of thing
> ?

As an example, 10 % random cells with distances to 10% of those:

r.mapcalc residuals=map1-map2
r.univar residuals

n: 4720
minimum: -0.682179
maximum: 6.86607
range: 7.54825
mean: 0.0994347
standard deviation: 0.418746
variance: 0.175348
variation coefficient: 421.127 %
sum: 469.332

So doesn't seem like a huge difference, but when I compare the
distribution of distances with a kolmogorov-smirnov test they are
statistically different. I'm not a statistician though so maybe this
isn't a valid measure.

Any comments?

Joel