Long processing times in hodo_plot.py due to cfradial write and read. Solution, plus possible bug in pyart code?

[lcarlaw]

Writing and subsequently reading a cfradial version of the pyart radar object takes a very long time. In benchmark testing, this averaged near 35-40% of the total hodograph processing time.

The pyart.retrieve.vad_browning can evidently be run using sweeps extracted from the raw/native pyart radar object (initial data format is MSG31 binary), simply by converting radar_1sweep = ncrad.extract_sweeps([idx]) to radar_1sweep = radar.extract_sweeps([idx]), and removing the write_cfradial and read_cfradial calls. A try/except block must be added though, to handle sweeps in which not insufficient data is available which otherwise would result in a ValueError.

One minor item of note: somewhat inexplicably, there are subtle/minor differences in the data when comparing what's in the original radar object and what's read in by pyart.io.read_cfradial. This differences are clearly introduced either by write_cfradial or read_cfradial. Regardless, the differences are slight (see images below), but probably not enough to justify the significant time increase to utilize cfradial data.

Example of differences caused entirely by pyart.io.write_cfradial and/or pyart.io.read_cfradial:

Hodograph using cfradial data:

Hodograph using original pyart.io.read object:

[lcarlaw]

Commit https://github.com/tjturnage/cloud-radar-server/commit/a18887aec47f609b46140a48c71c9c0f2afcf31f turned off the write/read cfradial steps in favor of using the original pyart.io.read object given the substantial speed increases.

[lcarlaw]

This is not an active bug, or enhancement area. Simply maintaining this for reference purposes in the future.