Run down of model biases.....

[ColdFront77]

What do the different models mean, how they have done with past storms, why they indicate certain things, which ones are better and for what reasons... this way we all understand in one concise thread.

[Guest]

OK well I do not quite understand the question..

I believe you are asking for a breakdown of each particular model and how they run, and then the difference between an amateur acting like he/she knows something to a professional who works for a living doing this.

[Stormchaser16]

Well..... im nowhere near a professional, but i personally dont find it the least bit difficult to look at a pretty colored map and analyze it lol.....

To be serious though, i can understand the models, and what they are saying, its just something that i have grown used to over the past few years

[garyb]

Try this link http://www.nhc.noaa.gov/aboutmodels.shtml

[Guest]

Tom like anything else in life, in you take the time to study, listen, and observe you will learn invaluable info. Me personally I learned much of my model info from a certain few Mets who in detail broke down numerous model runs to the T. I would and still do save the analysis and study it myself trying to pick up more and more intel as I go. Also in this day in age a simple google search gives you hours upon hours of valuable intel to decipher through.

Another important aspect to me personally with this subject is taking a model hook line and sinker everytime out. I've seen many people burned by this method over and over again. This is a harder task to me because you have to look outside of what the model is telling you and break things down for yourself via upper air patterns ETC. Always a work in progress even for pro mets IMO.

[ColdFront77]

OK well I do not quite understand the question..

I believe you are asking for a breakdown of each particular model and how they run, and then the difference between an amateur acting like he/she knows something to a professional who works for a living doing this.

Brian, that's pretty much what I am referring to.

Well..... im nowhere near a professional, but i personally dont find it the least bit difficult to look at a pretty colored map and analyze it lol.....

To be serious though, i can understand the models, and what they are saying, its just something that i have grown used to over the past few years

Each model acts differently and some are more "accurate" than others depending on several factors.

I see nothing too difficult about looking at lines on a map... but it is always said that there are differences and having them in one thread would be easier,
wouldn't you say?

Try this link http://www.nhc.noaa.gov/aboutmodels.shtml

I thought we could have a discussion about this subject in one thread; apparently not.

Tom like anything else in life, in you take the time to study, listen, and observe you will learn invaluable info. Me personally I learned much of my model info from a certain few Mets who in detail broke down numerous model runs to the T. I would and still do save the analysis and study it myself trying to pick up more and more intel as I go. Also in this day in age a simple google search gives you hours upon hours of valuable intel to decipher through.

Another important aspect to me personally with this subject is taking a model hook line and sinker everytime out. I've seen many people burned by this method over and over again. This is a harder task to me because you have to look outside of what the model is telling you and break things down for yourself via upper air patterns ETC. Always a work in progress even for pro mets IMO.

I can't disagree with you, Md. I do search for things online. There are several questions about models on the board. So, why not bring it up?

Again, I felt that we could have a discussion on this. I guess it won't work and is another one of my topics that isn't going to go anywhere.

[Guest]

Me personally, question #2 I have been studying met literally since I was 6, and I did study for a couple years at PSC in new hampshire. so i guess I am in the catagory as a avid amateur heh.

Here is my take: anyone, no matter how much or little you know, can look at a site like barometer bob and go to the little models link and look at a few UKmet and GFS chart and say 'OH well the GFS says this and that'. That is not forecasting and definitly has no credibility.


So anyway, (ill try to keep this short) - believe that as a forecaster, you need to do many things to somewhat accurately predict potential changing patterns, and their effect on tropical weather systems. First of all, the initialization. the data that was collected and input into the model and how each run will be affected. Then comparison. How can you compare so and sos model versus the previous run. and the run before that. what has changed? what went wrong on a trough that didn't dig as far as it should have. the subtle differences. You can look at the GFS ensembles for instance. take the 500mb and surface data for each run. (the ensembles I believe have 20 different runs), each with it's own initialization slightly different in different regions, different strengths of pressure, changes in jet stream strength, etc etc which will create different setups. Then the 'mean' is generally shown on websites, the average of the entire suite. The most important part is knowing the biases in the models, such as the numerical models like UKmet, GFs, NGM, Canadian ECMWF etc. Each one has it's own way of anazlying and interpeting data. if youre unfamiliar with a certain models characteristics and it's weaknesses then chances are you are crapshooting when comparing it to another model. Each one has it's own personality, to put it simply. historical data, real time data, and etc all play into things and knowing it's traits is important for figuring out what it's worth.

I could go on and on but i'm rambling. Im sure a pro can better explain, anyway!

[ColdFront77]

I'd like to make it clear, I am bringing this topic up for everyone interested in reading and posting to this thread, not just myself.

I have been interested in and follow the weather for many years, but the tropical weather forecast models that we see on this site are only grouped together once in awhile and each discussed individually. It can get confusing even for someone knowing meteorology. If that sounds odd, oh well.

As for the website that was posted. First off, I have been there several times and secondly... I find it a bit strange how people can work, read this board and do other things in their lives and read all this meteorology literature versus me being online frequently because I generally have nothing else to do.

[opera ghost]

I think what he's looking for is a run down on the biases...

From what I've picked up over the last two seasons- I think the

GFS is biased left

BAM does best with a medium level system (low cat hurricanes?)

B(something something)D works with deep level systems (I've always assumed this was major hurricanes)

(something something) L does best with low level systems (tropical storms and depressions?)

LBAR deals best with systems that stay far south and don't recurve

And I'm not even sure how much of that is actully correct. Wish I could be more helpful.

[Guest]

hehe..well good try anyway

[ColdFront77]

Then it takes time to remember what is and is not the general consistency with the models.

[Guest]

Tom, good read if you've never seen this courtesy of DT.

http://www.wxrisk.com/hurricane/FrcstTe ... rules.html

[ColdFront77]

Thanks! .. .. ..

Hopefully I am not coming across as ignorant of meteorology. I shouldn't post my thoughts, etc.

[Guest]

Model analysis of current TS Frances..

http://www.hurricanealley.net/Storms/06LMDLSFA.html

[Pebbles]

Tom, good read if you've never seen this courtesy of DT.

http://www.wxrisk.com/hurricane/FrcstTe ... rules.html

Thanks for sharing!

[Wthrman13]

It depends on how much detail you want to get into. I can give you a little overview of how the dynamical models work, since that is my area of study.

The various models used for tropical cyclone track and intensity forecasting vary by many orders of magnitude (if one can use such a measure) in complexity, with the dynamical models being by far the most complex. By "dynamical" I mean actual numerical weather prediciton (NWP) models that are based on a set (typically about 7) of nonlinear partial differential equations, called the "governing" or "primitive" equations, that describe the most important atmospheric motions and processes. These equations are impossible, even for the most gifted mathematicians out there, to solve in a pure analytical sense. What I mean by that is that I cannot take those equations, isolate each variable (i.e. temperature, pressure, density, water content, wind speed and direction, etc.) and write down an equation for each variable that is a solution to the differential equations. The best that can be done is to approximate the solution to the equations numerically by breaking them up into discrete intervals in space and time and crunching the resulting algebraic equations (which can be solved analytically) for each grid interval in both space and time. This requires enormous computing power. Also, there are many, many, many, many ways to "break up", or discretize, the governing equations, some of which are better than others in different circumstances. When designing an NWP model, very careful attention must be paid to how the discretization is done, because if done incorrectly, your model will blow up in your face by giving incredibly wrong predictions of the various variables. Even with a good numerical scheme, there are an incredible amount of assumptions that go into writing out the original equations in the first place, some of which are dubious at best with our current knowledge of the physics of the atmosphere. In particular, our knowledge of how cloud and ice particles form and interact in clouds is very limited at this time, and our equations that go into the NWP models are crude and limited. Turbulence and how to account for it is also a huge issue. So is surface interactions. All of these are absolutely essential to forecasting many atmospheric motions, particularly the smaller-scale ones such as thunderstorms. Another major issue for NWP is the initialization problem. That is how do I get a highly irregular and sparse set of observations of the atmosphere (i.e. upper-air soundings, surface station obs, radar data, etc.) into a model framework that in fact needs values of the observed variables at every grid point? There exist many techniques for doing this in a consistent, statistically robust manner, and again, some are better than others. This is another area of intense ongoing research. As I've said many times before, it's amazing the dynamical models work at all, given all the assumptions and sometimes outright hand-waving that goes into them.

Now, as far as how this pertains to tropical prediction, one thing that many people fail to understand is that, at least as far as the dynamical models go, there is nothing really different about the tropics as far as the model is concerned. The model doesn't get south of 30 N and say, "Oh wait, I'm in the tropics now, I gotta change to a new set of equations." Well, actually, that is an oversimplification, because there are some differences in the equations depending on what exactly is going on at a given location, but by and large the governing equations are considered to be adequate to explain all major atmospheric motions, including tropical and mid-latitude motions. Hidden inside these equations are solutions (sometimes called "modes") that correspond to mid-latitude cyclones, to tropical cyclones, to large-scale planetary waves, to thunderstorm squall lines, to tornadoes, and yes, even sound waves. In fact, we wouldn't want a different set of equations for different regions of the atmosphere. What we want is to be able to describe all major atmospheric phenomena with one concise set of equations, and that is what the various dynamical models, both global and regional, try to do.

Now, different models, especially those developed by different agencies, are likely to differ in many respects. These include, but are not limited to:

1) What form of the original governing equations to use. Certain forms are better for describing different scales and classes of atmospheric flows. If you are developing a model that only deals with the lowest 1 km or so of the atmosphere near the ground (called the boundary layer), you are likely to use a set of equations called the Boussinesq equations. Similarly, if your model is designed to forecast very large scale motions in the atmosphere, you can get away with a much simplified version of the governing equations called the quas-geostrophic equations. Nowadays, since computing power is so prolific, most NWP models out there use the full set of equations with as few assumptions as possible and are capable of simulating a wide-range of flows. For example, the ARPS model, which I work with, is capable of simulating everything from large-scale troughs to tornadoes (a huge range of spatial and temporal scale).

2) The method of initialization. Some techniques, which I won't try to describe as this post is already way too long, include optimal interpolation, 3d-variational (3DVAR), 4d-variational (4DVAR), and Kalman filtering (KF).

3) Convective, cloud, turbulence, and surface process parameterizations. These are the poorest understood areas of weather modeling right now, as I mentioned earlier, and using different techniques can cause a huge difference in the result of the forecast.

Since each agency treats these issues differently, it's no surprise that the different models have different ideas of what is going to happen. An example of course is currently with Frances, where the GFS has a much different track than the UKMET, both high-performance global dynamical models. It's difficult to ascertain at times what is causing these differences, and to identify biases and quirks of the various models, but the differences I outlined above are largely responsible for the spread in the forecasts of the various models out there. The GFS, for example, used to have a problem with spinning up nearly every tropical disturbance into a tropical cyclone. This problem has been largely alleviated and fixed. The goal, of course, is to develop more and more robust models with much less forecast spread, especially in hard-to-predict regimes such as tropical cyclones and thunderstorms. As we understand turbulence, microphysics, and convection more in the coming years, this should improve.

I could write more, but this post is already huge, maybe I'll discuss more in another post later.

Dan

[x-y-no]

Thanks Dan, great answer.

If you choose to post more, maybe you could talk about the "baroclinic" and "barometric" track-prediction models like GFDL, the BAM models, LBAR, etc.

First of all, what do "baroclinic" and "barometric" mean in this context. I have looked up the words, but was unable to make sense of how the definitions I found applied to models.

Second, as I understand it these models use the dynamic GFS model as their basis, and apply additional calculations to get a better understanding of cyclone structure and track. Am I more or less right? And can you explain in at least a qualitative sense what they are doing?

Thanks,
Jan

[CaluWxBill]

I think it is Barotropic but either way. I am not sure what they mean when they talk about the models. But a barotropic scenario is one in which the atmospheric temperature profile stacks with the height profiles. In a baroclinic scenario a strong difference in temperatures set up and they set up a pattern where temperature patterns cross the height patterns in the atmosphere, creating areas of temperature advection. generally storms in the midlatitudes are a response to baroclinic conditions creating strong boundaries (cold fronts, and warm fronts.)

[x-y-no]

I think it is Barotropic but either way.

Oops ... yes, my bad. don't know how my fingers typed "barometric", lol.

I am not sure what they mean when they talk about the models. But a barotropic scenario is one in which the atmospheric temperature profile stacks with the height profiles. In a baroclinic scenario a strong difference in temperatures set up and they set up a pattern where temperature patterns cross the height patterns in the atmosphere, creating areas of temperature advection. generally storms in the midlatitudes are a response to baroclinic conditions creating strong boundaries (cold fronts, and warm fronts.)

Yeah, that's the definitions I found, and I couldn't figure out how that applies to models.

[CaluWxBill]

I think it is Barotropic but either way.

Oops ... yes, my bad. don't know how my fingers typed "barometric", lol.

I am not sure what they mean when they talk about the models. But a barotropic scenario is one in which the atmospheric temperature profile stacks with the height profiles. In a baroclinic scenario a strong difference in temperatures set up and they set up a pattern where temperature patterns cross the height patterns in the atmosphere, creating areas of temperature advection. generally storms in the midlatitudes are a response to baroclinic conditions creating strong boundaries (cold fronts, and warm fronts.)

Yeah, that's the definitions I found, and I couldn't figure out how that applies to models.

Well I would presume that most models are created to handle cyclogenesis in a baroclinically induced system, in the midlatitudes. Tropical systems are a little different in that they do not really form from differences in temperature, rather in convective systems near the ITCZ, models do not have the resolution to handle these systems, often times making them much weaker than they would be. The GFDL seems to be set up to forecast for strength, whereas AVN and UKMet and other global models, really only forecast well the steering conditions of storms and probably do not forecast intensity well. hopefully somebody can better explain 3D tropical modeling to me.