Re: another online prediction tool

Jerry
 

That is kind of what I did with my GPSL predictions only manually.  I always ran 6 predictions, 60K, 90K & 105K burst  1100FPM & 850FPM Asc/desc.  Then for good measure I would add the habhub prediction as a test.



I have been thinking about adding a way in my tool to automatically run the multiple runs but never got around to it.  My tool never really caught on so haven't spent much time on it.

 
Jerry Gable
Balloon Flight Prediction tools
http://www.s3research.com



From: "'L. Paul Verhage' nearsys@... [GPSL]"
To: GPSL@...
Sent: Tuesday, March 28, 2017 7:40 AM
Subject: Re: [GPSL] another online prediction tool

 
So I wonder if we could create a good landing zone prediction by varying the ascent rate and burst altitudes by plus or minus some percent. Say 10%. Then use a combination of these variations to map out a recovery ellipse (I assume it would an ellipse shape with the major axis aligned along the overall flight path).
On Mar 28, 2017 8:35 AM, "Jerry Gable jerrygable@... [GPSL]" <GPSL-noreply@...> wrote:
[Attachment(s) from Jerry Gable included below]

To me it looks like their tool takes in variances in the weather data but my experience has been human caused variances such as fill and burst altitude. 

Here is my correlation from GPSL last year.  After the flights I ran a prediction with the actual ascent, decent, and burst.  I used the preflight weather data and they all came out pretty close.



 
Jerry Gable
Balloon Flight Prediction tools
http://www.s3research.com



From: "Mark Conner mconner1@... [GPSL]" <GPSL-noreply@...>
To:
Cc: GPSL list <GPSL@...>
Sent: Monday, March 27, 2017 1:29 PM
Subject: Re: [GPSL] another online prediction tool

 
Not only are U and V covariant with each other to some degree, but there is covariance with height.  Each wind level is not independent of the levels adjacent to it - in fact they are highly correlated.  If you have winds at 180/25 at 250 mb, they will be very close to that value at 240 and 260 mb.  

Below is a heat map for my planned Saturday launch.  I think it underestimates the left/right uncertainty in the track, and probably the distance as well.  It certainly underestimates the overall uncertainty in using a 120-hour forecast versus, say, a 24-hour forecast.

Don't get me wrong, doing a Monte Carlo approach is a good way to get a peek at uncertainty.  But I'm not sure if this is a full measure of the overall uncertainty in the landing site prediction.

73 de Mark N9XTN


Inline image 1

On Mon, Mar 27, 2017 at 11:40 AM, Mark Conner <mconner1@...> wrote:
The model treats it as U and V.  The model is on a Cartesian grid but "knows" it's on a sphere (the physics account for that).

It's more a statistical exercise in the wind errors.  Is it better to treat U and V errors independently, or is there covariance?  Does that covariance change as the vector magnitudes change?  Winds aren't truly random, so Gaussian statistics may not apply.  

So is it better to assume that U has a standard deviation of n m/sec and V has a SD of y m/sec, or is it better to assume that direction has a standard deviation of q degrees and speed a SD or r m/sec?  The "r" part would not be Gaussian because it could not be negative, so some other distribution would have to be assumed.

Something I'd like to explore in the literature more.  Someday.

- Mark


On Mon, Mar 27, 2017 at 11:17 AM, 'L. Paul Verhage' nearsys@... [GPSL] <GPSL-noreply@...> wrote:


Interesting, they describe the transition at Hank's Knee as the Drag Crisis Band. The paper looks very interesting in my brief and cursory look. It appears the Reynold's Number of the balloon makes a gradual descrease during the entire ascent.
Mark's comment makes me wonder now how the GFS model generates wind profiles. Is it in the north-south direction and the east-west direction or does it generate speed and direction directly. I guess this could be referred to as generating wind profiles in Cartesian or Polar coordinates. As we will recall from high school, each can be converted into the other. But which is easier to compute for supercomputers generating the GFS model (enquiring minds want to know)?
;)
Still, all kidding aside, the paper and the questions it brings to mind are interesting. I'll find some time to read it in more detail. For example, I'd like to learn what percentage of variation it puts into the data for each prediction.
On Mar 27, 2017 10:03 AM, "Mark Conner mconner1@... [GPSL]" <GPSL-noreply@...> wrote:


The full paper describing the methodology is here:

The Reynolds number transition, mentioned in an EOSS presentation at GPSL several years ago, is modeled here.  For others with balloon flight forecasting tools, there are some concepts here that could be leveraged.

The treatment of wind uncertainty is interesting.  Most models, including this one, treat the uncertainty independently in the u (east/west) and v (north/south) vector components.  I've always thought, without anything to really back it up, that uncertainties should be handled in direction and speed space and then taken back to u/v for computations.  But I don't know if that will produce numerically different results in the end.

73 de Mark N9XTN

On Sun, Mar 26, 2017 at 4:43 PM, 'L. Paul Verhage' nearsys@... [GPSL] <GPSL-noreply@...> wrote:


This is much quicker than running several predictions with various input parameters. I think this form of modeling a landing zone can be called the Monte Carlo method.
On Mar 26, 2017 10:42 AM, "Mark Conner mconner1@... [GPSL]" <GPSL-noreply@...> wrote:



This one is interesting in that it will run numerous predictions with variations in the launch parameters and return a heat map showing the likely landing zone.  I did one for our NSTAR 17-A launch from Lincoln scheduled for next Saturday.

Inline image 1

Around 50 simulations seems to be enough to generate a reasonable map.  I tried 50 and 150 for this flight and the heat map did not change very much.

73 de Mark N9XTN


















Join GPSL@groups.io to automatically receive all group messages.