Plutonium core based meteodrones use HIGH RADIATION GAMMA IMPULSE for provokation HUGE water condensed process in previosly still atmosphere...meansured, please), GAMMA LEVEL near typical "supercell" & understand this message.

An excellent question, near and dear to my heart! It's also the same question I asked the engineers working on the Blue Waters supercomputer before it was fully built - I wanted to know, and their own answer was they didn't know! The short answer to your question about bit-perfect reproducibility is "yes" but only under specific circumstances. The CM1 model I use is compiled into a binary executable from source code (like all models that run on supercomputers). If I use the same exact binary I should get the same exact results, down to the very last bit, so bit-identical. I have verified this. However, a while back I tried to reproduce a simulation using a newly compiled version of the model and immediately saw that the solution was very slightly different, and the differences got larger over time (this is expected to happen with models like CM1 which simulate highly nonlinear phenomena). So what happened? Updates to the compiler and updates to the libraries on the supercomputer. The source code did not change, but the results were different in the least significant bits of data, and these small changes amplified over time, like the "butterfly flaps its wings" scenario. In science, reproducibility is considered monumentally important. You may have heard news stories where social science or medical studies cannot be reproduced (usually they involve human subjects). This has been called the "reproducibility crisis." In my world of cloud modeling, I am no longer so worried about this because any given simulation is just one out of a nearly infinite number of outcomes. I can always create a new simulation, and often times I get a very similar result (i.e., a big fat EF5 tornado that lasts a long time) even when I recompile the model or run the model on a different machine. I am always careful to point these things out, and the atmospheric science community I think generally understands this. It's also why the focus from specific individual simulations has changed to *ensembles* of simulations where you look at a bunch of simulations in slightly different environments/configurations and do statistics on them. In ensembles, it is hoped that the statistics will not change much even if each individual run may have changed. But it is especially challenging to do ensembles at the scale I typically run my simulations at, since my whole approach is to push the hardware very hard. So I kind of consider each simulation to be similar to an actual storm chase - with a storm chase you only get one shot on a storm. For super big simulations (like the 10 meter El Reno 2011 simulation I did on Blue Waters) I will never get that bit-perfect result if I run it again because the machine that I ran it on no longer exists, and there will be very slight differences in mathematical operations on a different machine.

@@LeighOrfsThunderstormResearch - Thanks for your reply. Alas, it doesn't *_unequivocally_* answer the question, for at least two reasons. One of them you mentioned-namely, you haven't had the opportunity to run the simulation on the same hardware twice. If the way the program runs is dependent on idiosyncrasies that vary from one machine to another, then it would be impossible to answer the question with any certainty. Let me back up and tell you why I've asked the question. I'm not concerned about a "reproducibility crisis". That's a concern only for physicists and others who deal with simple systems. You're dealing with a complex system; the Newtonian paradigm doesn't apply. The question is, have you truly succeeded in modeling a complex system? Now, if you run the program multiple times on the *_same_* hardware and you get exactly the same result every time, you haven't created a successful model of a complex system. From my perspective, it's a "boring" result. That's not a criticism, and it certainly isn't in any way a deprecation of the massive amount of ingenious work that you've obviously done to create these awesome and unique simulations. Nor is an assertion that these sims are not a giant leap forward in understanding tornadogenesis. I rather expect the opposite is true...

​@@LeighOrfsThunderstormResearch​ I might be wrong on this, but I don't suppose these kinds of models are great candidates for being run on GPU based supercomputers? I would guess there's too much interdependence between different data points for that to be efficient, but maybe I'm wrong on that. Do you think newer systems are moving in the direction that would let you realistically run multiple iterations of the same model (maybe with slight noise added to the input data or model constants) in the next few years?

I love the fact that some of the stuff he's given acronyms to relates to some of my favorite animes

I'd love to see an flcl vorticity

Plutonium core based meteodrones use HIGH RADIATION GAMMA IMPULSE for provokation HUGE water condensed process in previosly still atmosphere...meansured, please), GAMMA LEVEL near typical "supercell" & understand this message.

Yup, you're not the first person to make the connection! Very cool stuff!

You beat me to this comment by an hour lol. Skip talbot told me to check this video out after I noted the unique structure. I recall watching it a while back, but it was cool to revisit after being mesmerized by reeds footage. Being over an open dust field created a perfect real life example of the simulation! Great stuff!

I was about to say the same thing and realised I'm 8 days late to the party! Was watching Ryan Hall's stream when it happened and had no idea how visibly striking the tornado actually was until the next day when Timmer uploaded the footage. Watching this video for the first time today having already seen Reed's video and the parallels are astonishing!

That footage Reed got is some of the most hauntingly beautiful video I’ve ever seen. I was absolutely mesmerised by its structure.

Take a look at a typical undergraduate meteorology degree curriculum. 3 semesters of calculus, linear algebra, differential equations, 2 semesters of physics, 1 semester of basic chemistry. And of course all of the meteorology classes that are sequenced to work with the prereqs. It's been a while since I taught undergrad but I'm pretty sure that about covers it. If you are into computers, excellent, the more programming you know the better. One thing I have painfully realized is I am no 'true' software engineer, I'm a scientist who does a lot of scientific programming. Good luck.