[GRASS-user] curve sketching

[Jose Gomez-Dans]

Hi Martin,

On 6/5/07, Martin Wegmann <wegmann at biozentrum.uni-wuerzburg.de> wrote:
> - point of global maximum (already in r.series -> max_raster)
> - point of global minimum (already in r.series -> min_raster)
> - turning point  (Wendepunkt)
> - point of max./min. slope (e.g. growing season, senescence)
> - information about unimodal/bimodal etc. phenology (no idea how to add this)

I work with similar time series data (it looks as if you're interested
in monitoring phenology, probably based on some veg. index or some
other satellite derived biophysical variable). The problem with these
series (and I guess with many other data of this kind) is that the
data is very noisy (changes in solar illumination, sensor geometry,
atmospheric effects...). If you filter out the noise, you could do
away with temporal resolution. Over homogenous regions, you can filter
spatially, thus reducing your spatial resolution. Either way,
resolution degradation. You can also fit curves to your time series
(double logistic functions and so on), and use the fit parameters to
infer onset of senescence, budburst dates, etc.

What I am trying to get at is that this very application and data
source dependent. It would be very hard to code something which is
generally useful. What I do is to export the time series into Python,
and process them further there. The results so far are very good for
crop phenology monitoring using MODIS data. The scipy python module
has a lot of very useful functionality to do any of the stuff I
mentioned above.

My two cents :)