1 hour ago, blingthinger said:

 

Guess? There's no guessing here. I've already linked many NOAA links showing that it's CFD. Which one are you struggling to understand?

 

 

just because its using states in finite volumes doesn't mean its CFD

 

 

1 hour ago, mSparks said:

just because its using states in finite volumes doesn't mean its CFD

I apologize that was interpreted like that. It wasn't meant to be implied that finite volume is always CFD. The wikipedia page I linked earlier about finite volumes uses examples that are not explicitly related to fluid dynamics. My highlighting of "finite volume" is more trying to connect the dots for you and anyone else reading this thread. FV does happen to be a discretization method of the Navier Stokes equations and is extremely common. When FV is found in a sentence along with NS (Euler eqn is a subset of NS), it's going to be describing a CFD algorithm.

What else are you confused about?

Edited January 23, 20251 yr by blingthinger

10 minutes ago, blingthinger said:

I apologize that was interpreted like that. It wasn't meant to be implied. The wikipedia page I linked earlier about finite volumes uses examples that are not explicitly related to fluid dynamics. My highlighting of "finite volume" is more trying to connect the dots for you and anyone else reading this thread. FV does happen to be a discretization method of the Navier Stokes equations and is extremely common. When FV is found in a sentence along with NS (Euler eqn is a subset of NS), it's going to be describing a CFD algorithm.

What else are you confused about?

the FVs give you probabilities.

The "normal" way to turn probabilities into forecasts is DTMC

From

to

I'm confused how you think CFD helps get you a temperature reading from a weather balloon sent up 12 hours ago to a temperature estimate for a weather balloon now, and in 3/6/9 hours from now

Edited January 23, 20251 yr by mSparks

https://www.cfd-online.com/Wiki/Finite_volume

10 hours ago, blingthinger said:

 When FV is found in a sentence along with NS (Euler eqn is a subset of NS), it's going to be describing a CFD algorithm.

Definitely correct.

12 hours ago, mSparks said:

the FVs give you probabilities.

hahaha This isn't anywhere near correct which means you aren't reading anything that's been linked before or are too word not allowed to continue the conversation. I'm going with the former. 

We have yet to see a single NOAA documentation link backing your nonsense up. We've clearly entered the phase of msparks $-it posts :)

2 hours ago, Murmur said:

Definitely correct.

I don't know why I didn't click the top link on this page a long time ago, but this page contains quite a few documents describing the FV3 algorithm at the core of GFS/GDAS.

https://www.gfdl.noaa.gov/fv3/fv3-documentation-and-references/

Very first link points to this:

https://repository.library.noaa.gov/view/noaa/30725

Equation 3.6 is the Navier Stokes momentum component in its final form before they chunk it up into the finite volume format a few pages later. What's also fun is that if you word search for 'CFD', it appears no less than 10 times! Gasp!!!

Edited January 23, 20251 yr by blingthinger

Give them all a bit more time and you'll get the AI-massaged ensembles too...

3 minutes ago, jcomm said:

Give them all a bit more time and you'll get the AI-massaged ensembles too...

I seriously thought that's what generated his kinetic energy backscatter and Kalman filter post a couple days ago but that turned out to be legit...so this time it was tough to tell. Baby steps I guess!

5 minutes ago, blingthinger said:

I seriously thought that's what generated his kinetic energy backscatter and Kalman filter post a couple days ago but that turned out to be legit...so this time it was tough to tell. Baby steps I guess!

at work, as I've mentioned, I use "only" ECMWF and ARPÈGE with runs adapted for Continental Portugal. But ECMWF is starting to walk on the "AI ensemble road" 🙂

We're all AI-ized... Bummer! When I started working at the Metoffice 30 yrs ago I was comming from AI, and by that time only one of my colleagues was already well aware and oriented towards THIS future ... Now I'm just too old 🙂

 

5 minutes ago, jcomm said:

Bummer

Ah! I thought you were referring to the past posts in this thread. You're actually talking about models like this:

https://arstechnica.com/science/2024/12/googles-deepmind-tackles-weather-forecasting-with-great-performance/

It's wild to see it predict more accurately than the CFD model(s). I'm curious....why do you say 'bummer' to this one?

Edited January 23, 20251 yr by blingthinger

48 minutes ago, blingthinger said:

it appears no less than 10 times! Gasp!!!

with the last time it being mentioned is when they explain

https://doi.org/10.1002/wcc.318

is more important.

A reasonable link.... to a locked article about climate (not weather), none of whose authors are associated with NOAA and its prediction tools.

I never said stats models are never used in weather prediction. I merely said GFS is a CFD model at the core. It most certainly does use statistical data reductions in some of the submodel components, but the fundamental algorithm is not statistical. It's physical: conservation of mass, momentum, and energy.

As for how a weather balloon data set can be used in the CFD?

I said awhile ago that it's used as the initial condition for the solution. They'll set the grid points at that region to the values recorded by the balloon all the way up in altitude. They'll apply some statistical method (Kalman?) to create an average of the data that was recorded over the period of the balloon flight (hr or so?), at each grid point the balloon came close to, as well as blend with neighboring data points recorded dozens of miles away. Then the CFD starts iterating on the future 3 hr forecast and beyond with that data-based distribution (seems like they call it a global analysis ensemble pdf page 6, step 9) as the starting point.

Edited January 23, 20251 yr by blingthinger

15 minutes ago, blingthinger said:

A reasonable link

its from your link, the last reference it has to CFD

Quote

the role of both forms of diffusion in engineering CFD applications. In recent years numerical diffusion and anti-diffusive noise generation has played an important role in the development of “stochastic models”, which is a fancy name for a model in which random perturbations are injected into some felds (Franzke et al., 2015). 

 

16 minutes ago, blingthinger said:

Then the CFD starts iterating on the future 3 hr forecast

Like I said, no need to guess, link LOC, 

 

19 minutes ago, blingthinger said:

but the fundamental algorithm is not statistical.

DTMC and its peers  is no more "stastical" than CFD solvers. the difference is the input and output, which is probabilistic and doesnt need to assume a functional form to move from one cycle to the next.

using a CFD solver for bits like wind makes plenty of sense, but everything else? temps, humidity, cloud coverage and formation, precip? I dont see how you can use CFD for any kind of global weather, at least not with any kind accuracy, all it will tell you is the weather is chaotic, and we dont need a model to know that.

Me gathering probabilistic stats for XP weather after deciding to use the almanac reading for Louisiana earlier this week.

Edited January 23, 20251 yr by blingthinger