Confidence Levels

This was entirely too confusing in it's previous incarnation. I've simplified the explanations somewhat.

Short Term Only. All models are in close agreement. Discrepancies =< 10 miles
Short Term Only. Predictions point to an area =< 15

Long and Medium Term predictions with very similar flight vectors and relatively minor variations in distance.

Short Range predictions pointing to an area less than 20 miles in circumference.

Long and/or Medium Term Sequential Predictions in agreement and pointing to the same general vector.

Short Range predictions are separated by less than 30 miles

Either too early a prediction to rely on or if close to launch day, predictions are wildly erratic and thus not to be trusted.

Post Burst Prediction Confidence Levels

The post burst predictions are something else entirely. These predictions use the winds encountered during the ascent phase of the flight. The shorter the flight in both time and distance, the more accurate this prediction will be, the longer in either time or distance will push this prediction away from perfection. However, as a general rule, this prediction usually places the balloon within a mile or so of its actual touchdown. I would always rate the confidence in that prediction HIGH.

Deviation by Example

If you wish to be convinced of how a deviation from expected operating parameters can impact a prediction, check out the maps showing the flight tracks of EOSS-105 and EOSS-106. Both balloons were extremely similar in flight configuration and they were launched within 46 minutes of each other. However, one developed a pin hole leak leading to a very slow ascent to a float at around 60K feet, the other flew a pretty rapid ascent to 98K feet. The first balloon was cut down, the second ascended to burst. The result, they flew in the same direction but they landed 23 miles apart.

How Confidence is Acquired

I'm not a meteorologist, so I have to depend on others to give me confidence or the lack thereof in a prediction. Those other people are the authors of the meteorological models that I use to make predictions for our flights.

Basically I develop confidence in a prediction through two avenues.

First off, a prediction based a long time in the future does NOT have my confidence. There is only one model that will predict beyond 84 hours into the future. So, it is not possible to compare different models. On very rare occasions predictions from a week or more out actually agree somewhat with what happens on flight day. Much more often the difference is dramatic.

When we get close to flight day other prediction models come into effect.


Model Name Prognostication Window Increment Max Altitude
RUC 12 hours hourly ~54,000 ft. (100 HPa)
NAM 12 km 48 hours 3 hour ~ 68,000 ft. (50 HPa)
NAM 40 km 84 hours 3 hour ~ 68,000 ft. (50 HPa)
GFS short 0-84 hours 3 hour ~ 88,000 ft. (20 HPa)
GFS medium 0-180 hours 6 hour increments ~ 88,000 ft. (20 HPa)
GFS long 192-384 hours 12 hour increments ~ 88,000 ft. (20 HPa)
AFWA 15 km 48 hours 3 hour ~ 68,000 ft. (50 HPa)
AFWA 45 km 72 hours 3 hour ~ 68,000 ft. (50 HPa)

Prognostication Window

Below is a chart that shows the number of hours a given model can forecast into the future.


  1   2   3   4   5   6   7   8     16

Model                        Hours

12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192 204 ... 384
NAM 12 km                                      
NAM 40 km                                      
GFS short                                      
GFS medium                                      
GFS long                                      
AFWA 15 km                                      
AFWA 45 km                                      

note: the chart has been truncated between 204 hours and 384 hours to make it fit here on a web page.

As you can see, there are four basic models. However, within the sub-categories of several of the models there are differences. For instance, the NAM and AFWA models have distances associated with them. These distances are actually the size of the cells that are examined in the model to create predictions. The smaller the cell should theoretically render more accurate model predictions. Because much more data is being crunched this type of forecast is not run as far into the future and thus the shorter the prediction window.

I use the model that is most accurate and of the shortest length (window) using the assumption that this model should be the most accurate. However, I also like to get those winds at 88K feet. So, my preference is for the GFS if it is in agreement with one or more of the other models.

I rarely use the RUC. This is mostly because I think that a model that doesn't extend to at least 50 HPa in height is useless unless combined with another source of data to round out the prediction data. But, it may be more accurate than any of the others and best practices may suggest I use that blended with another source for the stratospheric level winds. Oops, I'm probably getting into ... more than you want to know.

Just a tiny bit of recent history. NAM used to be ETA, AFWA used to be AVN. I don't know if more than a simple cosmetic name change occurred for these models (algorithms tweaked).


Model Names expanded.

RUC = Rapid Update Cycle

NAM = North American Mesoscale

GFS = Global Forecast System

AFWA = Air Force Weather Agency

Models are limited by atmospheric pressure not altitude. However, I list the altitudes as most folks aren't really conversant with what altitude each pressure reading generally occurs. The important thing to note is that the altitudes are approximate as high pressure raises max altitude and low pressure lowers it.