Closed kongdd closed 4 months ago
this is an interesting idea, how do you expect to pass this information ? as an extra band ?
@mgravey and @agathearchidoit,
Could give some explaination about the parameter extrapolate_flag
?
https://github.com/open-geocomputing/OpenEarthEngineLibrary/blob/5f82c48e5dde59c4728e465360c65df28acf037b/ImageCollection/SavatskyGolayFilter#L51-L54
In theory, extrapolate_flag
should always be zero.
https://code.earthengine.google.com/a1c4e3359e1c4fdc351238c74becfbbc
There are only three conditions:
t
in the range of [-nt, nt]
, extrapolate = 0
;t
in the range of [-nt, 0]
, extrapolate = 0
;t
in the range of [0, nt]
, extrapolate = 0
;In my R package phenofit, I implemented the weighted SG.
But it is a little bit complex to translate. I will try whether it is possible.
The extrapolation flag tell you when all conditioning points are on a single side. Usually it's a sign that it would be very wrong :) But why do you imagine that it should be always at 0 ?, typically the first and last point of a series are not. But it can happen even in the middle of a series if it's to parse compared to the windows.
Based on Wikipedia it doesn't look complicated.
You still didn't answer to my question! how do you expect to pass this information, as a band ? One for each variable ? As a feature for each image ?
saveAllJoin
will always include the current image (i.e., dT = 0). Hence, extrapolate
does not work as expected.The estimation collection can be very different of the inputCollection, if you work with MODIS, you are probably both at a daily basis, but with Landsat or Sentinel it allow to get estimation on a constant intervals, even if the input is not.
Sorry wrong button
Points with different quality, should give different weights. Any idea and any plan to implement this feature?
Thanks