Forecast in the past?

All issues/questions about EMS v3.4 package, please ask here.
Post Reply
Larf
Posts: 16
Joined: Tue Dec 02, 2014 10:03 am
Location: Hamburg

Forecast in the past?

Post by Larf » Tue Dec 02, 2014 12:57 pm

Hi,
maybe its just me (and I have to live with me) but I'm desperately seeking for a hint in the tutorial, forums and roberts descriptions if the model is producing a forecast, using the datasets or making an analysis.

I have downloaded two weeks of cfsr data for my site and I do have a comparison measurement. The time series correlate. Thats good news. But isn't the model generating the timeseries on my spot using the input timeseries and the model? Or is it taking the initial conditions of the two weeks and forecasting from there on?? Wouldn't make sense, if two weeks of data are available.

So why is it called "Initial Forecast Hour" and "Final Forecast Hour"? Whats the word "forecast" for?"
And where in the tutorial can I read anything about that?

I hope the answer is at least a little bit tricky!

meteoadriatic
Posts: 1565
Joined: Wed Aug 19, 2009 10:05 am

Re: Forecast in the past?

Post by meteoadriatic » Tue Dec 02, 2014 1:50 pm

I'm not sure if I understood your question correctly. But...

You have for example two weeks of cfsr data and want to run WRF model over those two weeks? No problem, but you have mutliple options to do that.

Option A is to use first grib from your cfsr data series and it will be initial conditions, and all other gribs will be only for boundary conditions of your WRF domain. This means that you will run two weeks at once, initialize model at first day and it will finish at last day. During model's run time, it will take all other data just to feed boundaries of your domain with characteristcs of airmasses that enters from outside into your domain during that time.

Option B is to split those two weeks, for example into 14 pieces, and then you will run 14 WRF runs one after other in 1 day lenght. Or do any other splitting, for example 2 runs of 7 days. Then, WRF will be initialized with cfsr data at each time you start new instance of model's run.

I do not particularly recommend any of those two approaches over other, until I know all details about project. Both approaches has some advantages and some drawbacks over other.

Larf
Posts: 16
Joined: Tue Dec 02, 2014 10:03 am
Location: Hamburg

Re: Forecast in the past?

Post by Larf » Tue Dec 02, 2014 3:30 pm

Thank you very much, meto.
So where do I get to make this choice? And where can I read more about it.

My case:
Study and verification of the wrf model using met mast data for correlation. After enough verification and understanding hopefully using the model for wind farm modeling especially in complex terrain.

Thank you again and good night
Larf

meteoadriatic
Posts: 1565
Joined: Wed Aug 19, 2009 10:05 am

Re: Forecast in the past?

Post by meteoadriatic » Tue Dec 02, 2014 4:50 pm

How do you run WRF model at all? Did you managed to actually run anything with your downloaded data or you got unresolvable problem? I mean, just for test, can you manage to run any test reanalysis using downloaded data?

Larf
Posts: 16
Joined: Tue Dec 02, 2014 10:03 am
Location: Hamburg

Re: Forecast in the past?

Post by Larf » Wed Dec 03, 2014 9:35 am

I installed wrfems and followeed thems guide. So I generated the domain and folders using domwiz and successfully downloaed 6 hour values of cfsr data from October 15th through 28th 2010 in order to compare to a met mast. I did that before at another location and gained acceptable results. The time series corresponds with the measurement at site. I hope this wasn't bare chance!

meteoadriatic
Posts: 1565
Joined: Wed Aug 19, 2009 10:05 am

Re: Forecast in the past?

Post by meteoadriatic » Wed Dec 03, 2014 1:41 pm

Good, can you now use ems_prep followed by ems_run and ems_post using correct options to generate wrf forecast from October 15th 00z up to October 16th 00z?

For example something like, exectue within domain directory:

Code: Select all

ems_prep --date 20101015 --cycle 00 --length 24 --dset cfsr
then

Code: Select all

ems_run
finally

Code: Select all

ems_post
If you have multiple domains also pass swithes for them. That will generate 24 hours forecast.

Then you can go on and run another 24h forecast:

Code: Select all

ems_prep --date 20101016 --cycle 00 --length 24 --dset cfsr

Code: Select all

ems_run

Code: Select all

ems_post
And so on until end of your period. Try that first and let us know if you find a problem.

Larf
Posts: 16
Joined: Tue Dec 02, 2014 10:03 am
Location: Hamburg

Re: Forecast in the past?

Post by Larf » Wed Dec 03, 2014 3:40 pm

Hi Meteoadriatic
My machine is currently running the domain for the 2 weeks setup. I passed the line
ems_prep --dset cfsr --analysis --date 20101015 --cycle 00 --length 336
I think that takes at least until tomorrow.

So my guess is that I chose the option A: with the first dataset as initial condition and the further sets as boundary conditions. Good. My original question and what confused me was the term "forecast", because in my opinion it is a mesoscale analysis at a certain area refined by nests for micrositing and more precise results.

Because you stated: "Try that first and let us know if you find a problem." --- Actually everything went fine! It was just for understanding the approach.

BTW: Thanks. I apriciate your effort.

Next run, I could try to apply your suggestions using
ems_prep --dset cfsr --date 20101015 --cycle 00 --length 24
ems_prep --dset cfsr --date 20101016 --cycle 00 --length 24
...
and compare the results

But that will take a while

meteoadriatic
Posts: 1565
Joined: Wed Aug 19, 2009 10:05 am

Re: Forecast in the past?

Post by meteoadriatic » Wed Dec 03, 2014 3:44 pm

Great, let us know how it is going!

Post Reply

Who is online

Users browsing this forum: Baidu [Spider] and 2 guests