[R-sig-Geo] Trend significance bfast

[Guillaume Clair]

Hi,

I'm using the function (and package) bfast to work on some NDVI MODIS datas
in Kenya.

I modified the original function in order to extract directly the only
datas that I need, i.e. the trend component (Tt) and the dates of
breakpoints in trend (bp.Vt)

I'd like to process the results in order to get the slopes (this I managed
since it is easy math), and also very importantly I would like to get some
significance of the trend, like it is generally possible to get with the
parameter ANOVA = T in the plot.bfast function. We can see an exemple of
the result HERE <http://fr.tinypic.com/r/doo3l0/9>.

I have two separate questions, the first one is critical, the second is
more out of curiosity in order to understand something that I don't
understand for now :

- in the source code of the function plot.bfast (
https://github.com/cran/bfast/blob/d8d853096d628b62251c6d302319fa95c14e0206/R/plot.bfast.R),
the lines 36 and 37 are processing the ANOVA parameter. Could someone
explain me what is the seasonal() function which is used, and which I
cannot find anywhere ? What I would like to do is to adapt/create my own
script in order to get the same information directly from my own processed
datas (only Tt and bp.Vt), since I cannot process it again by lack of
computing power. I hope the absence of the ci.Vt won't be a problem...

- does someone know why sometime the ANOVA parameter doesn't work ? And
instead return an error

Error in out$bp.Wt$datatsp : $ operator is invalid for atomic vectors

with a plotting just partially made...

Thank you all for your attention.

Best regards.

Guillaume

R Version : 3.3.1

Platform : x86_64-w64-mingw32

	[[alternative HTML version deleted]]