The Search for our Missing Balloon: Closing in on the Landing Zone

In late May, Jared Sabater of Soleil Multimedia and I launched a high altitude balloon from south Chapel Hill, North Carolina. The balloon was carrying three cameras to capture spectacular high resolution video images of North Carolina from 20 miles in the air. We also expect to see the black sky of near space and a slightly curved horizon. However, the satellite tracker package fell off immediately after launch, and the balloon — cameras in tow — disappeared into the sky.

We set off to the expected landing zone — Harnett County, North Carolina. Needless to say, the search made looking for a needle in a haystack sound easy, and we returned home with nothing. I posted a desperate plea on my blog (read it here) and crossed my fingers, hoping someone would find the payload and get in contact with me. A reporter from the Daily Record in Dunn, North Carolina, came across my blog post and wrote a story about it.  A day later, I got a call from a woman living in Coats, North Carolina.

“I saw your balloon,” she said.

I didn’t believe her at first.  The balloon was supposed to pop in the stratosphere, not come down intact.  She said she saw it at sunset, a full twelve hours after launch — I figured it should have been way out in the Atlantic by then.  But as she described what she saw, I realized that there were only two possibilities:
1.  She saw our balloon.

2.  Someone else launched a high altitude balloon that just happened to come down in Coats.

The probability of #2 is vanishingly small — smaller even than #1, so I was forced to conclude that, in fact, our balloon was somewhere by Coats.  Jared, my friend Xiao, and I drive out to Coats to talk with the witness.  She was wonderfully nice and amazingly observant.  We stood by the window from which she saw the balloon, and she started telling me what she saw.  First of all, her pastor had also seen it crossing a road to the north, so we had a bearing.  Second, she said it had something sticking off to the side.  Third, it had remained in the same place for about an hour.

I didn’t know what to make of the second and third statements.  Balloons don’t fly with things sticking sideways, that makes them unbalanced and they tend to rotate so that whatever it is sticks straight down.  Also, even when the wind is imperceptible at the Earth’s surface, it’s strong enough above the surface to move the balloon out of her viewing area in less than an hour.

That’s when I realized what she’d seen:  Our balloon had already landed by the time she glanced out her window!!

It was simple in retrospect.  By the time the witness looked out her window and saw the balloon, the payload was on the ground or snagged in a tree.  The balloon envelope was still buoyant, hanging above the landing site like a giant flag — pulled slightly sideways and hooked over by the wind (hence the thing sticking out) and more or less stationary (why she was able to see it for an hour in the same place).

Here’s what we know right now:

The balloon envelope is cream to reddish colored and probably up to 20′ across.  It’s likely shredded into bits due to UV light and wind action.  The payload is 40′ from the envelope in a red lunch box with cameras attached.  The envelope is likely draped over a tree, and the payload is probably in the upper branches of another tree.

The area visible to the witness starts at 35.389756 N latitude, 78.638441 W longitude, and extends due east.  We can define sight boundaries by using buildings and trees that obscure her view to the east and the west (see figures below).

We searched several fields in the sight line and didn’t find anything, so the payload’s likely in forested areas.

Another witness saw the balloon crossing Red Hill Church Road at very low elevation, heading towards the first witness’ house, so that road defines a hard eastern boundary to the search area.

The first witness says the balloon was west of Black River – this provides another search area boundary, but with less confidence.

Elevation profiles (see below) and distances to the Black River make it even more unlikely that the witness could have seen the balloon on the ground if it were in the Black River valley.

Here are a series of maps of the search area:

view_areaThe red triangle defines the region visible from the first witness’ house, with the east boundary defined from the road which the balloon crossed per the second witness’ description.  Green polygons enclosed regions we were able to search on our first trip out to Coats.  The pale blue arc is 1 mile from the first witness’ house.  It’s hard to imagine being able to make out as much detail as she reported if the balloon is beyond this circle.

Elevation profiles along the north and south sight lines are available here:



These profiles suggest that the Black River area would be difficult to see from the first witness’ location.


The four yellow polygons show the most likely landing zones based on the analyses described above.  Areas A and B are by far the most probable locations for the payload, based on the witness’ statements and distance from where she saw it.  Any search should carefully investigate these regions.  Area C is also possible, but rather less probable because it is significantly below (and thus probably invisible from) the witness’ house.  Also, it’s quite far away, which would make the details she described hard to see.  Finally Area D cannot be eliminated, but it is the least likely due to elevation and distance.

Our next step is to go out and search.  If you’d like to join, or if you have some other information for us, let us know!


Extracting North American Mesoscale (NAM) Model Data with rNOMADS using DODS

The (relatively) recent upgrade of rNOMADS to version 2.0.0 allows cross platform support for downloading atmospheric and oceanic operational model data into R via the GrADS-DODS system.  However, users have been experiencing difficulty using DODS to get North American Mesoscale (NAM) real time data.  The script below shows an example of getting the NAM 12z model for North America using DODS.  Below the script and the figure, I’ll talk about why getting NAM data via DODS is different than, say, GFS.

#Get data for NAM subsets using DODS
#Individual NAM models are listed separately in GRIB but all together in DODS.
#So we get the dates of the NAM runs first, then we have to sort out which "type" of NAM we want.

all.nam.dates <- GetDODSDates("nam", request.sleep = 1)

#Here, we pull the model runs.  This is when we can get individual models out.
#Get the most recent one: <- GetDODSModelRuns(tail(all.nam.dates$url, 1))

#For example, let's get North America "nam_na" model, 12z model run.
namna.model.runs <-$[grepl("nam_na_12z",$]

#We can get info on what that model contains by inspecting the output from this:
info <- GetDODSModelRunInfo(tail(all.nam.dates$url, 1), tail(namna.model.runs, 1))

#Get data on surface solar flux across North America <- DODSGrab(tail(all.nam.dates$url, 1), tail(namna.model.runs, 1), "csdsfsfc", 0, c(0,706), c(0, 285))

#Now we can plot it with rgl 
#First, set the "missing data" values to -1 (or whatever)

#Credit to Dr. David Forrest for this display code:
valScrubbed[valScrubbed==9.999e+20]<- -1

Solar radiation flux over North America from the NAM NA model.

Solar radiation flux over North America from the NAM NA model.

The reason we can’t just get a specific NAM subset model using (for example)

GetDODSDates("nam_na", request.sleep = 1)

is because the DODS server is set up differently than the GRIB server for NAM. It turns out that all the NAM models are stored together in one directory (in GRIB, each one has its own directory), and so we actually have to sort out individual models at the model run level, not the date level. So the way the above script deals with this is:
1. Figure out all dates when NAM was run.
2. Figure out the model runs for a specific date.
3. Pull out the specific NAM model subset you want (e.g. nam_na, nam_ak, whatever).
4. Get the model data.

If you have questions, don’t hesitate to contact me:

rNOMADS 2.0.1 released

The latest version of rNOMADS is now available on CRAN.  This update resolves several minor bugs and one major one involving multiple variable/level selections when using the ModelGrid function.  I have also added support for two more models on NOMADS:  Climate Forecast System Flux Products and Climate Forecast System 3D Pressure Products.  This brings the total number of real time model products supported by rNOMADS to 57.  A quick test of the Climate Forecast System Flux Products produced the following image of surface albedo for 1200 GMT on July 4th, 2014:

Albedo of the Earth's surface at 1200 GMT, July 4th, 2014.

Albedo of the Earth’s surface at 1200 GMT, July 4th, 2014.

In other news, data curators at the University Corporation for Atmospheric Research (UCAR) are currently investigating rNOMADS as an open source software package to read their GRIB archive.  UCAR hosts a series of model archive data, some of which is freely available to the public.  Their specific interest is in using rNOMADS to read their archive of the FNL model.  They have gotten wgrib2 to work on MAC OS X, allowing rNOMADS GRIB functionality to be extended to Macintosh machines.