Closed martin-rdz closed 2 years ago
@martin-rdz Thanks for your informative comments.
Under current version, the mask for particle depolarization ratio is rather simple with only filtering data points with relative uncertainty larger than 30%. See below
But apparently, your method is more suitable with more features remained.
I will have a test with your method first. And hopefully implement it soon...
@ZPYin Thanks for taking care 👍 Looking forward for version 2.0
I was looking into the python version of the plotting routine, but didn't grasp where the filtering was done.
/lib/polly_general_func_lib/pollyxt_display_retrieving.py
Would it be possible to also filter the data that goes into the netcdf file? Or at least introduce a flag? Otherwise the end-users would have to calculate the molecular profiles thereself.
Some improvements have been implemented in commit: c57dbdc
I increased the upper limit of the relative uncertainty for filtering noise par-depol points (from 30% to 60%), instead of implementing your criteria. Because, the uncertainty caused by particle backscatter, molecular depol and signal shot noise have been taken into account, based on the error propagation equation. Therefore, your criteria 2 can be realized by introducing a threshold for uncertainty of par-depol.
Your criteria 1 and 3 were adapted to better contrain the results, but with a different threshold. (see below)
To help end-users to use the par-depol in a better way, the uncertainty of par-depol was stored in the **profile**
files. Besides, molecule depolarization ratio
was saved in the attributes of parDepol***
. see below,
Data visualization requires a lot of controls, in order to increase the readability of each figure. However, the framework of Picasso v2.0
is dedicated for data processing instead of data visualization, which makes it hard to be maintained and improved for features of data visualization. That's one of reason that I want to introduce Picasso v3.0.
I will summarize issues associated with data visualization and fix them together in the new version.
@Moritz-TROPOS , wanted to work on a substantial way to calculate the depol for the Polly systems. Thus, the issue might be reopened using another name.....
@Moritz-TROPOS , wanted to work on a substantial way to calculate the depol for the Polly systems. Thus, the issue might be reopened using another name.....
Welcome~~~ It's reopened now, feel free to add comments.
With regard to code development, please refer to issue #12 and #15. 🍻
any further comments or can it be closed or moved to discussions? @Moritz-TROPOS @HolgerPollyNet
Problem
The particle depolarization ratio estimate is rather noisy and plotted incompletely. Example from the Punta Arenas measurements is shown in the left two plots below, two features are marked at 2.0 and 5.5 km. The retrieved data as saved in the netcdf is shown on the right (blue curve in the 3rd plot and orange curve in the 4th plot).
The exploding values in the particle depol are caused by a singularity in the formula for low ratios of molecular_bsc/aersol_backscatter. When the aerosol backscatter is smaller than than the molecular backscatter, this behaviour becomes very pronounced (everything left of green curve in the plot below)
Proposed solution
Include a filter based on the volume depol, the molecular depol and the ratio molecular_bsc/aerosol bsc. An example is shown in the 4th plot (first figure, green line).
Several steps seem to work rather well, without requiring additional smoothing (only tested for Klett retrieved 532 backscatter so far)
volume depol < molecular depol
tomolecular_depol
>40
[optional, but already reduces noise quite a bit]0.7
Formula for step 3 is with the coefficients
a,b
depending on the desired particle depol thresholdExample implementation
In python, as I am not fluent in matlab and not sure how masking is done in the processing chain. Sorry for not including it on my own.
Likely the two thresholds (
40
from step 2 and0.7
from step 3) should be subject of the config file.