Let the user provide custom weight into pyresample

[mraspaud]

From lavergne...@gmail.com on November 25, 2013 14:15:49

Hei pyresample team,

I am seating at the pytroll workshop, and would like a new feature:

Users could specify an array of weights (same dimension as lat/lon/data), and this weight would multiply the weight computed from distance (through the fov function, either _custom() or _gauss()).

Two use cases:

1. the weight could be given as 1./var(data) : the inverse of the uncertainty in each data point (as variance);
2. the weight could be computed on fov observation time, with more weight towards 12utc than at 0utc and 23:59utc. A resulting daily composite image would then be "sharpened towards 12utc", wrt to a daily composite without any weighting.

I'll start implementing right now :)
Thomas

Original issue: http://code.google.com/p/pyresample/issues/detail?id=12

[mraspaud]

From lavergne...@gmail.com on November 26, 2013 00:41:37

OK. I have a version I would like to share with you, Esben. Can you take a look?

I implemented this as an extra keyword parameter custom_weight=None to resample_gauss() and resample_custom(). This new feature should have no impact for resample_nn(). If custom_weight is None (default), I checked that the testsuite runs smoothly. I have also created 2 test cases for the new feature, and will add more today.

Please find attached kd_tree.py with the new feature, and test_kd_tree.py. This version is tentatively named 1.1.1.

I also attach an animated gif that cycles between 3 images of the "average sensing time" (0-24 hours) when doing a daily composite of SSM/I swath data in the Arctic. That uses resample_gauss(). The loop is going through "pyresample-1.1.0", "pyresample-1.1.1_withCustomWeight", and "pyresample-1.1.1_withoutCustomWeight".

In that case, the custom_weight I use is a "hat" function that gives a weight of 1 at sensing_time 12utc, and a weight of 0 at sensing_time 00utc and 24utc. The idea is to get a daily averaged image that is sharpened towards 12utc.

So we see that "pyresample-1.1.0" is the same as "pyresample-1.1.1_withoutCusomWeights" (frame 1 and 3), whille pyresample-1.1.1_withCustomWeights (frame 2) does what we want (gives less weight to the late 22utc swath).

What do you think, how do we proceed from here?

Thomas

Attachment: kd_tree.py test_kd_tree.py

[mraspaud]

From lavergne...@gmail.com on November 26, 2013 00:50:48

Posting the 3 frames separately, since animated gif didn't pass.

Attachment: test_pr111_times_frame1.png test_pr111_times_frame2.png test_pr111_times_frame3.png

[mraspaud]

@TomLav Is this implemented now ?

[TomLav]

Short answer: no.
Longer answer: maybe. I have been thinking lately that pyresample has maybe all what is needed to achieve this feature without touching the code. I must give it a try.