Identifying phenomenon trends

[martijn at scanner.frw.uva.nl]

You might want to run r.mapcalc to produce a truly binary map (just X and i), 
then use a neighbourhood operation with diversity as its criterion. This will
select border cells on either side of the cliff coast, which you can thin out
with r.thin and vectorise with r.line. If you have a lot of Xi borders that
are irrelevant to the coastal movement, eliminate them beforehand by fiddling
the region boundaries and/or by masking them (creating the mask with r.digit).