Illuminated Supercells
Вставка
- Опубліковано 1 сер 2023
- I found a few new buttons on Paraview the other day. I subsequently went on a 'render bender' and this is the final result. Last year the software engineers at Kitware, authors of Paraview, included some GPU-native new lighting code that does things I've mostly only seen done with hard to use expensive proprietary software. Coupled with an NVIDIA Titan RTX GPU that had been spending far too much time idling, I was able to make this entire video in only a couple days. I was able to render up to 4 movies in parallel and the GPU only used up about half its available memory.
Some technical details:
This is a CM1 numerical model simulation in a highly idealized environment. Paraview, the visualization software creating each PNG frame that makes up each movie, is reading ZFP-compressed CF-compliant NetCDF files created by my own software (that I call LOFS). I found the best performance by saving individual files for each variable at each time rather than putting all the variables in one file at each time. I am using Volume rendering with shading turned on, with the Global Illumination Reach parameter set at about 0.2, Volumetric Scattering Blending around 1.1, and Volume Anisotropy around 0.0 (settings varied between movies).
All software used in running, processing, and visualizing this simulation is open source. My sincere thanks to all programmers who were involved in these projects!
Regarding the supercell simulation:
Just recently I began using a different microphysics option (NSSL microphysics, authored by Ted Mansell) in CM1. What you see here is a large eddy simulation of a supercell (75 meter grid spacing, which is actually a bit coarse for most of my work, which has gone up to 10 meters) using NSSL microphysics, with data saved every 2 seconds, starting at time=0. The storm is initiated with a "warm bubble" perturbation. The model domain is 120km by 120km by 30km with visualized data extending up to 22 km.
Lots of interesting physics is revealed in this simulation (showing supercells split, following by the dominant right mover). A few highlights:
1. The AACP behavior by both the right and left mover. Just, wow. There is graupel in those plumes in addition to cloud ice. [NOTE: After talking with Ted Mansell, this excess of graupel/hail is likely to go away with newer versions of his code, which I am going to be using soon.]
2. The relationship between hydrometeors (rain, hail, graupel) and the cold pool (temperature perturbations, outflow patterns). Clearly, some downdrafts are dominated by thermodynamics, while others are dominated by dynamics.
3. The 'waterfall effect' of hail and graupel at the edge of the cloud
4. Tornadogenesis is associated with some explosive updraft growth at the tropopause
5. Isn't it cool how environmental vorticity gets reorganized (and quickly depleted) in the environment surrounding the rapidly growing updraft near the beginning of the simulation
These types of movies reinforce my long-held belief that visualization, when done right, can foster true scientific discovery that is often missed by theory and observations.
This work was supported by NSF grant AGS-2114757 and a Texas Advanced Computing Center (TACC) subaward from NSF grant OAC-2139536. - Наука та технологія
Wowed again! The detail in the crashing waves behind the overshooting top, the AACP and everything else is beautiful even at 75 meter resolution. And we got to see it do the splits! Hopefully we get to hear Dr. Orf narrate his thoughts of these simulations.
Hello pecks hank, and congratulations on 1 million subscribers! I’ve been watching you for years and you sparked my interest in tornadoes.
Yeah I am way past due for a little narration. You know I have lots of thoughts haha.....
The emergence of mammatus clouds at the bottom right of the screen at 1:35 Is absolutely amazing
These simulations are about to change everything. Absolutely revolutionary and a wonderful breakthrough with atmospheric science. Bravo Orf, Bravo.
Any more corroborating photos of actual supercell formation from Pecos Hank? It would be great to see a split screen time linked to a closely matching event.
Good point, with this new volume rendering approach I finally have shadows and effects that look actually somewhat real, so Hank & I need to put our heads together again I think...
This is absolutely amazing. Who would have ever thought we would come to a point where we would have the technology to view atmospheric physics in this high of resolution, let alone have the people who are intelligent enough to program this level of information. Incredible!
Always cool when I get a new video notification from you.
Fluid dynamics is so sick!
obsoletely stunning
That beast at the end
17:10 would go hard as a desktop wallpaper
Yet another incredible simulation! Good job!
Basically I am just subscribed because of your collab with @PecosHank but again and again these videos of your research turn out to be fascinating. Keep up the good work.
_Differential equations describe the world!_
Confirmed we live in a lava lamp simulation.
There's a lot happening in all of these simulations but the detail and complexity of vort magnitude at the end is truly bonkers.
Wait until I do the same simulation at 30 meters.... hehehe....
Mesmeric
Absolutely incredible!!
i was pretty sure i was the only one to remember that graupel was a thing, took quite a few searches especially considering that my memory of it was spelled "grovel". stunning viz work all around. i'm curious if you saw the recent awesome footage from Reed Timmer and how it must validate (or refine, or challenge) your understand of tornadic simulation. have you done a computation based on, and compared to, an actual super cell event? left side simulation, right side satellite photo....that style of video?
more people should see this..
Man you may be on to something, I believe it has to do with individual gusts Leigh, to create the vortices.
What's the difference between a successful tornado and an unsuccessful tornado? All your models have been able to create a tornado but I think I would like to see a model of an unsuccessful formation of a tornado. See the differences. Or weal tornados to see the differences between a weak tornado and a massive strong and intense tornado.
Very cool! would also make a good screen saver
17:36 great moment
Fascinating stuff! Will that super computer also tell you precisely when the Vikings will implode this season? ;)
Siiiiiiiick 😲
You mentioned the data is plotted for every two seconds of simulation and I'm wondering what the actual temporal scale each simulation represents....that is what does each second of simulation represent in real time for each view/mode? BTW, the left updraft split reminds me of Ted Mansell's early simulation of the what I think was the Dimmit Tx supercell well over a decade ago, IIRC.
The model time step (the 'temporal resolution' of the model itself) for this simulation was 0.5 seconds, but I'm saving data every 2 model seconds, so there are three "frames missing" between each of the frames you see here (4k 30fps). There's not much point to save data at much of a higher temporal resolution for views this far out, but I've done it before when studying the tornado itself.
Very interesting I was wondering if the "puff" that comes out of the main updraft around 7100-7300s caused the end of the cell and tornado, it looks like it almost splits the updraft in two and then at around 8000-8200s it reaches the anvil and totally moves the center of the overshooting top, any ideas?
It honestly looks like the storm starts to lean back up until that point and then the shot of sinking air to the base of the storm killed the structure
Dr. Orf - That is so impressive. I've worked with developing GPU parallelled compute nodes for general applications and databases, but not for fluid dynamics. How many points are in this sim? wow excellent work!!
Simulation (run on 128 Frontera CPU nodes - each node has 56 cores): nx=1600 ny=1600 nz=397 (1 billion grid zones). The files rendered were nx=1296 ny=996 nz=292 (377 Mpoints). Really not a "big" simulation/rendering... there will be much more to come and some will be at much higher resolution. I have access to much larger GPUs on Frontera, which also runs Paraview, so really the sky is the limit (time to render poster-sized images LOL)
@@LeighOrfsThunderstormResearch Incredible!! That's the Cornell System?
what are the skinny streaks of green to orange that occasionally fall out of the cumulus clouds? cool as hell btw
I think they’re tornado tracks
I think you're referring to the graupel view 2? Those graupel "plumes" must be originating from microphysical processes involving say hail, snow, etc... that I'm not rendering, so they look like they just appear out of thin air...
okay thanks
Am I seeing it incorrectly, or does the first thunderstorm simulation seem to be rotating anti-cyclonically?
Correct. The left moving supercell rotates anticyclonically. In the southern hemisphere, an anticyclonic supercell would rotate counter clockwise, the same direction our as our northern hemisphere cyclonic supercells.
Looks like oil 🛢
what CPU were you using?
For the simulation, it's Frontera, which is manufactured by Dell. The rendering was done on 8 year old hardware basically... /proc/cpuinfo says "Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz" (10 cores). The CPU mostly just handles I/O, data uncompression, and sending data to the GPU which does the "real work". It also helps to have a PB of storage on the local network!
should have used an rtx 4090
Oh crap. I'll delete this and start over. LOL.
@@LeighOrfsThunderstormResearch lol on a related note, if you can add tornadoes to minecraft, ill buy you an rtx 4090.
@@shawnthesheep3720 I will keep that in mind. FYI Shaun The Sheep seasons 1 and 2 is the best television I've ever watched, hands down.
@@LeighOrfsThunderstormResearch thats funny. Im a huge weather nerd btw love your work