Archiving experimental files that were used to explore some crazy PDF diagnosis methods. Much of this is sporatic LLM generated code and reports. These two methods were tested and did work well when using LES input to model the PDF, but were extemely noisy and unstable in free running clubb. The 'parcel tracking' one performs a mixing-length-like calculation to evolve a persistent 3rd gaussian, then computes a residual ADG1 pair to balance everything out and adhere to clubbs predicted moments. The second one (dubbed trivariate even though they're technically all trivariate) is a 2 Gaussian method that allows for many more degrees of freedom in how we place Gaussian centers and allocate covariance - it does this in part by taking into account rt and thl skewness, which require calculating rtp3 and thlp3. This trivate actually does really well when modelled using LES data, but it's heavy reliance on xp3 terms makes it extremely noisy and bad in clubb - this is why this code also includes a much more sophisticated (and barely tested) advance_xp3_module.F90, as an attempt to make this less worse for the trivariate pdf.
(details / githubweb)
Fixing bug in advance_xp3 code. This is BFB because advance_xp3 is hardcoded to false.
(details / githubweb)
