[GRASSLIST:4507] Re: filter out extrem values

[Thomas Dewez]

Malte,

On a conceptual level, you may want to filter your data with a
standardizing filtering system.

The thing to do is interpolating your dem/ bathymetric model at first.
Create an average elevation map and a standard deviation map with a 3x3,
5x5 convolution window. Then you compute the difference between the
original depth and the averaged surface divided by the standard deviation.
Outliers are the points whose score is larger than 1.96 (threshold of a
gaussian distribution at 95% probability, if I am not mistaken). The
assumption under this is that the elevation inside your moving window
follow a normal distribution. Look into a table if the threshold of 1.96
was too small and ditched too many points.

r.neighbors may do the trick for creating the average and stdev images
while r.mapcalc will cover the "if" function and the image
difference and ratio computation. 

Does this correspond to what you are after?

Thomas

On Mon, 16 Sep 2002, Malte Martin wrote:

> Hi!
> 
> I got 35000 points where I measured the depths of a lake. Now I have to
> filter out the extrem values (maybe by moving average?). 
> 
> I want to interpolate the surface of the lake bottom.
> 
> 
> I'm not quite shure if s.windavg is what I'm looking for or shall I
> switch over to gstat? 
> 
> Thanks
> Malte
> 
>