How intelligent is r.cost?

[Conn Copas]

I am wanting to emulate an analysis problem from Tomlin's book on GIS and
cartographic modelling. Basically, it involves calculating the travel time to
a certain lake, where time is regarded as a function of slope, vegetation
density, etc. Each cell must ultimately be assigned a cumulative cost of
getting to the lake, where 'cost' is then interpreted as travel time. This
raises the issue of which point on the lake any other cell refers to. A
simplistic approach could use the centroid, but the man pages suggest that
GRASS permits one to input a layer which specifies multiple coordinates.

Questions : In the that case, can I simply use a map which contains 2
categories: lake and land? Or, do I need to create a map which defines the
perimeter of the lake only? I can vaguely see that this latter step requires
one to perform a neighbourhood analysis in which one inspects the attributes
of cells on each 'side' of a given cell, but exactly how would one do this in
practice? Once I have my map of multiple coordinates in a suitable form, can
I be sure that r.cost will select the 'cheapest' cell from that list when
computing cumulative cost for a given cell? Apologies for not having looked at
the algorithm first, but I was hoping someone might have a quick and easy
answer.

-- 
Conn V Copas
Loughborough University of Technology          tel : +44 509 263171  ext 4164
Computer-Human Interaction Research Centre     fax : +44 509 610815      
Leicestershire  LE11 3TU              e-mail - (Janet):C.V.Copas at uk.ac.lut   
G Britain                                      (Internet):C.V.Copas at lut.ac.uk