[GRASS-user] r.hants -how to determine frequency

[Rajat Nayak]

Thank you for the reply Veronica.
I shall try again with number of maps in one year as the base period and
with
varying FET.
Thank you again,
Regards

Rajat Nayak

On Mon, Feb 9, 2015 at 6:21 AM, Veronica Andreo <veroandreo at gmail.com>
wrote:

> Hi Rajat,
>
> > Dear All,
> >
> > I'm trying to reconstruct a cloud free NDVI time series using r.hants
> > module in GRASS.
> > I have a total of 318 files covering a period of 14 years (Feb 2000 - Dec
> > 2013).
> > I would like to know if there is any quantitative test to set the
> frequency
> > for this data range.
>
> Unfortunately there's no test to determine "nf" parameter in hants, but
> there's however a rule of thumb that works just fine... To consider at
> least 3 more frequencies than what you expect for your variable. For
> example, ndvi in a temperate area has 1 cycle per year, then you should use
> at least 4 nf.
>
> Quoting Markus Metz here "you need 4-6 cycles per year to accurately
> approximate NDVI with one peak per year. The additional cycles are
> needed to approximate seasonal differences in the increase or decrease
> of NDVI."
> The base period is by default the number of input maps. If you feed the
> algorithm with one year of data, you don't need to specify this parameter.
> If you, however, feed in the whole series of maps, you should use the
> number of maps you have in one year as base_period.
>
> > I used frequency values 4, with base period as 14, and frequencies of 28,
>
> > 42 and 56 without any value for base period. Other parameters set were,
> > high and low suppression flag, invalid data rejection range (-0.2 and
> 0.9),
> > Fit error tolerance of 0.05, and degree of over determinedness
> > as 155.
>
> AFAIK, range option is only for input data, it won't limit values in the
> output.
>
> The degree of overdetermination is how many extra maps you want to use to
> fit the curve. The algorithm per se uses 2*NF-1 valid data points to fit
> the curve. With dod you are specifying how many extra data points it should
> use (take into account you have a lot of missing values).
>
> For FET, I just tried different values ranging from 0.01 to 10, letting
> all the rest fixed.
>
> > However, none of the curves obtained match the expected distribution. (in
>
> > the study area there is cloud cover for 4 continuous months ).
>
> I worked with chlorophyll data in southern hemisphere and had the same
> problem... Maybe it will also happen to you that in those periods hants
> will overshoot and predict values outside the valid range of NDVI. You'll
> have to decide what to do with them. I fixed them with r.mapcalc (actually
> with t.rast.mapcalc).
>
> HTH :)
>
> Cheers,
>
> Vero
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20150209/6ceb2e5c/attachment.html>