Model’s Initial Conditions

A 2013 Steven Weiss lecture In SPC’s lecture series addresses numerical models’ initial conditions. He said that models may modify observational data in order to achieve the best possible Verification Score depending on the model’s purpose.

Is this something that is just as true today as it was then?

What are good sources for the most recent observed data? The SPC’s Mesoanalysis pages seem to be good, but they’re also being created by a model.

Is it even possible for a hobbyist to learn enough that they can recognize modeling modifications and therefore make their own forecasts and nowcasts better?
 

rdale

EF5
Mar 1, 2004
7,249
791
21
50
Lansing, MI
skywatch.org
It’s very much true. A bad set of initial conditions leads to a bad forecast. It’s okay that the model modifies things, but if it’s out of whack with reality then it is a sign :)

Use ASOS obs, upper-air soundings, and satellite/radar data to compare.
 
  • Like
Reactions: Michael Wilkinson

Jesse Risley

Staff member
Apr 12, 2006
2,159
586
11
40
Macomb, IL
www.tornadoguys.com
A 2013 Steven Weiss lecture In SPC’s lecture series addresses numerical models’ initial conditions. He said that models may modify observational data in order to achieve the best possible Verification Score depending on the model’s purpose.

Is this something that is just as true today as it was then?

What are good sources for the most recent observed data? The SPC’s Mesoanalysis pages seem to be good, but they’re also being created by a model.

Is it even possible for a hobbyist to learn enough that they can recognize modeling modifications and therefore make their own forecasts and nowcasts better?
RAP/UCAR: Upper-Air page: RAP Real-Time Weather
WPC Model diagnostics page: WPC's Model Diagnostics and Verification Page
 

Jeff Duda

Resident meteorological expert
Staff member
Supporter
Oct 7, 2008
3,317
2,062
21
Broomfield, CO
www.meteor.iastate.edu
A 2013 Steven Weiss lecture In SPC’s lecture series addresses numerical models’ initial conditions. He said that models may modify observational data in order to achieve the best possible Verification Score depending on the model’s purpose.
Your paraphrasing of this statement is backwards. The true statement is that observations are used to modify the ICs of a model run in order to improve them by "knocking" the first guess model state towards the observations. The process by which this is done is called "data assimilation." Data assimilation is as complicated as the weather model's dynamical core, and there are a lot of different approaches for it, each of which as advantanges and disadvantages. Some methods are more useful for assimilating observations representative of larger scale fields. Others work better on storm-scale fields.

Is this something that is just as true today as it was then?
Yes, if not more so. Modern day NWP forecasts are strongly dependent on the ability of a DA procedure to improve the first guess for the ICs for getting a good forecast.

What are good sources for the most recent observed data? The SPC’s Mesoanalysis pages seem to be good, but they’re also being created by a model.
Jesse provided some good sources for real-time observations. You should also look up individual state mesonets, because most states in the plains now have them (OK, KS, west TX, KS, CO, NE, IA all have state mesonets that can be displayed in real-time on their own individual websites).

The SPC mesoanalysis page shows data coming from SPC's internal "surface-OA" (sfcOA) software, which runs a simple form of data assimilation applied to recent surface obs only (i.e., does not reflect updated upper air conditions, even though there are few new ones provided each day outside of the classic 00/12 UTC rawinsondes). The background field for the sfcOA (i.e, the grid that it ingests) is a 1-hour RAP forecast, which enables the software to provide 3D atmospheric information even though there are few observations to fill the atmospheric space above the ground. The RAP has a 13-km grid spacing, which is why you see fuzzy gradients in products on the SPC page. My colleagues at ESRL/GSD have devised a much upgraded version of this idea, called 3DRTMA, similar to the existing operational RTMA. It is run in near-real time and graphics are available here: HRRR Model Fields - Experimental. It has been proposed as a replacement for SPC's sfcOA scheme, but politics and all are slowing the process of adapting it. Don't expect this system to become official use by SPC anytime soon (although you might see SPC run their sfcOA at 3 km sometime soon). FWIW, the 3DRTMA is based on the HRRR.

Is it even possible for a hobbyist to learn enough that they can recognize modeling modifications and therefore make their own forecasts and nowcasts better?
To be totally honest, unless you're willing to spend a lot of time researching this stuff (which is frequently being upgraded and improved), it will probably be difficult for a non-meteorologist to understand what all is going into these products enough to understand when they might not be performing well. Unfortunately that's a fact of life in a science field like this. Just do the best you can and ask questions when they come up, and hopefully someone will be able to shed some light for you.
 
Your paraphrasing of this statement is backwards. The true statement is that observations are used to modify the ICs of a model run in order to improve them by "knocking" the first guess model state towards the observations. The process by which this is done is called "data assimilation." Data assimilation is as complicated as the weather model's dynamical core, and there are a lot of different approaches for it, each of which as advantanges and disadvantages. Some methods are more useful for assimilating observations representative of larger scale fields. Others work better on storm-scale fields.



Yes, if not more so. Modern day NWP forecasts are strongly dependent on the ability of a DA procedure to improve the first guess for the ICs for getting a good forecast.



Jesse provided some good sources for real-time observations. You should also look up individual state mesonets, because most states in the plains now have them (OK, KS, west TX, KS, CO, NE, IA all have state mesonets that can be displayed in real-time on their own individual websites).

The SPC mesoanalysis page shows data coming from SPC's internal "surface-OA" (sfcOA) software, which runs a simple form of data assimilation applied to recent surface obs only (i.e., does not reflect updated upper air conditions, even though there are few new ones provided each day outside of the classic 00/12 UTC rawinsondes). The background field for the sfcOA (i.e, the grid that it ingests) is a 1-hour RAP forecast, which enables the software to provide 3D atmospheric information even though there are few observations to fill the atmospheric space above the ground. The RAP has a 13-km grid spacing, which is why you see fuzzy gradients in products on the SPC page. My colleagues at ESRL/GSD have devised a much upgraded version of this idea, called 3DRTMA, similar to the existing operational RTMA. It is run in near-real time and graphics are available here: HRRR Model Fields - Experimental. It has been proposed as a replacement for SPC's sfcOA scheme, but politics and all are slowing the process of adapting it. Don't expect this system to become official use by SPC anytime soon (although you might see SPC run their sfcOA at 3 km sometime soon). FWIW, the 3DRTMA is based on the HRRR.



To be totally honest, unless you're willing to spend a lot of time researching this stuff (which is frequently being upgraded and improved), it will probably be difficult for a non-meteorologist to understand what all is going into these products enough to understand when they might not be performing well. Unfortunately that's a fact of life in a science field like this. Just do the best you can and ask questions when they come up, and hopefully someone will be able to shed some light for you.
Thank you for the correction, suggested resources, and explanations.

It’s been very clear to me for a couple of years that there’s a tremendous gap between hobbyists and professionals. The more I learn, the more I’m impressed by how good the professionals are.

I’m amazed at the continuous development of tools like the HRRR. I’m excited to learn more about the 3D Real Time Mesoanalysis.

I appreciate the pros taking time to help the hobbyists. I’m grateful for all the online resources that allows a blue collar worker to expand their science knowledge a little. I’ve also noticed that by studying nearly every night after coming home from work, that I feel like I’m already on vacation. The value of that to someone like me is priceless.
 
  • Like
Reactions: JamesCaruso
Jul 5, 2009
1,110
940
21
Newtown, Pennsylvania
To be totally honest, unless you're willing to spend a lot of time researching this stuff (which is frequently being upgraded and improved), it will probably be difficult for a non-meteorologist to understand what all is going into these products enough to understand when they might not be performing well. Unfortunately that's a fact of life in a science field like this. Just do the best you can and ask questions when they come up, and hopefully someone will be able to shed some light for you.
I’m glad you said that Jeff. I have often felt like I *should* know more about model evolutions, the resulting biases, etc., but on the other hand I tried to not be too hard on myself, recognizing that it’s hard enough to stay up on the constant developments in my own profession, so how could I possibly expect to stay up on developments in meteorology, of which numerical weather prediction is only one specialized area of knowledge among many.

I’ve also noticed that by studying nearly every night after coming home from work, that I feel like I’m already on vacation. The value of that to someone like me is priceless.
I applaud you for doing this. With the uncertainty of whether I can even take my chase vacation this year, I tend to avoid studying much; I don’t want to get my hopes up for a trip that may not occur. It would be like looking through brochures about a resort you hope to visit, getting all pumped up for a trip that may not happen. Or like taking a cooking class when you know you may have to go on a diet before you can use any of the new recipes. I know, I’m weird that way, because what if the trip *does* happen? And even if I can’t use the knowledge this year, I can use it next year or the year after that. But like I said, I’m just weird that way. If I can’t chase, it’s easier to deal with it by just tuning it all out. Also hard to get motivated to spend the time, for something that may not happen....
 
I’m glad you said that Jeff. I have often felt like I *should* know more about model evolutions, the resulting biases, etc., but on the other hand I tried to not be too hard on myself, recognizing that it’s hard enough to stay up on the constant developments in my own profession, so how could I possibly expect to stay up on developments in meteorology, of which numerical weather prediction is only one specialized area of knowledge among many.



I applaud you for doing this. With the uncertainty of whether I can even take my chase vacation this year, I tend to avoid studying much; I don’t want to get my hopes up for a trip that may not occur. It would be like looking through brochures about a resort you hope to visit, getting all pumped up for a trip that may not happen. Or like taking a cooking class when you know you may have to go on a diet before you can use any of the new recipes. I know, I’m weird that way, because what if the trip *does* happen? And even if I can’t use the knowledge this year, I can use it next year or the year after that. But like I said, I’m just weird that way. If I can’t chase, it’s easier to deal with it by just tuning it all out. Also hard to get motivated to spend the time, for something that may not happen....
James,

You’re not weird. I think that way too in some other areas of my life. It would be better for me to change my way of thinking in those areas too, because every time I actually try to enjoy the path as much as the goal, it’s always better. There’s no risk. If I never reach the goal (due to either me or outside forces), it’s impossible for that failure to reach into the past and take away the fun and accomplishment that I had back then.

The fact that I’m able to do something like reading and studying (that was very difficult for me in high school) is giving me a better sense of confidence, and a craving to learn more. I fell kinda lucky that I enjoy something as complex as Meteorology, because I will always be able to pursue more knowledge about it.
 
Jul 5, 2009
1,110
940
21
Newtown, Pennsylvania
every time I actually try to enjoy the path as much as the goal, it’s always better.
I guess that's the key, I need to think about the journey and not just the destination, and try to enjoy the journey for its own sake, regardless of whether or not I ever get to the destination! I already chase that way - in other words, I do enjoy the process, not just the end result of a tornado or whatever - but I need to start viewing the studying in that way too, and enjoy it whether or not I actually get out to the Plains this year.
 
  • Like
Reactions: Michael Wilkinson