NSTAR INCIDENT ANALYSIS
(Getting close and then getting lucky)
Ralph Wallio, WØRPK  WØRPK  at  netINS.net

During the flight of NSTAR-01B one of two payloads separated from the parachute and fell 72,000ft to a crash-landing in rural western Iowa. Mark Conner, N9XTN, tells us a great deal about this adventure and eventual recovery of the damaged payload in his NSTAR web page at http://www.nstar.org/ and then http://www.nstar.org/nstar2000-01.html#01B. This discussion provides more details of methods used to understand when and where separation occurred and to estimate where the payload crash-landed. This was an interesting exercise regardless of the payload being discovered by a local resident and returned to Mark. Perhaps other high altitude balloonists will be interested in what can be learned from analysis of available evidence.

During the flight there was no significant indication that separation occurred. All APRS equipment was in the other payload so downlink telemetry continued without interruption. The separated payload contained a camcorder (which recorded the whole flight at 60 frames per second with time stamps right up to crash-landing) and a 70cm simplex repeater. The repeater continued to work through balloon burst, payload separation and as the payload fell free for 7 minutes. Mark and the recovery crew discovered the payload was missing when they located the remaining payload and parachute.

Back at home after a brief and fruitless search, Mark reviewed his PC recording of the simplex repeater and was able to determine with significant precision that the last simplex repeater transmission was at 14:50 UTC. This was our first piece of evidence that would be used to understand what happened. We now knew that the payload did not hit the ground before 14:50z.

Our second piece of evidence did not come so easily. Mark acquired partial APRS record files from Bill All, N3KKM, myself, himself and an APRS server (http://www.findu.com/). He then compiled an almost complete record of the flight which included time stamped navigation fixes every 30-seconds. Mark then loaded this compiled record to EXCEL and emailed the .XLS file to me for analysis. As received from Mark, this EXCEL file had columns for TIME, LATITUDE, LONGITUDE, COURSE, SPEED, ALTITUDE and VERTICAL VELOCITY (two values, one directly from the GPS receiver and the other indirectly from the receiver by calculations using altitudes and time stamps).  This is a small part of the file from just before burst (two columns to the extreme right are discussed below).



My first step was to attempt to determine when and where, latitude, longitude and altitude, the payload separated. The method was based on finding a significant reduction in descent rate when roughly half the payload weight fell away thereby reducing load on the parachute. This analysis took a bit of cut-and-try curve fitting by first building a fake wind file for input to WBALTRAK, a balloon track prediction program for windows offered as freeware by author Rick von Glahn, NØKKZ (http://www.eoss.org/wbaltrak/index.html). The fake wind file used arbitrary and constant winds all the way up to actual burst altitude (just to satisfy file format requirements) but with wind altitudes exactly the same as in the actual APRS data during descent. (At this point I was concerned about the accuracy of vertical, but not lateral, movement.)

Output from WBALTRAK showed bogus lateral movement caused by the placebo winds but it included valuable predictions of vertical velocity at the actual altitudes involved. As with normal balloon tracking exercises, these predictions were based on a sea level descent velocity estimate as calculated by the Descent Profile Calculator in WBALTRAK.  Output data showed estimates of higher descent velocities at upper altitudes (due to reduced atmospheric density) and velocity estimates gradually decreasing at lower altitudes all the way to landing (descent velocity in the CLIMB/DESCENT column).



Vertical velocity output from WBALTRAK was imported to the EXCEL file with altitudes matched so that across a row of the data are TIME, ALTITUDE (actual), DESCENT VELOCITY (actual), 780ft/min DESCENT VELOCITY (predicted) and 1050ft/min DESCENT VELOCITY (predicted). EXCEL charting was then used to show actual and predicted descent velocity curves.


The X-axis of this chart is GPS UTC time and the Y-axis is vertical velocity (values less than zero are descending). The chart starts with a few actual ascending vertical velocity data points before burst (roughly 1000ft/min) when, as it turns out, both payloads were still attached to the balloon and parachute. Right after burst the actual descent curve exactly matches a fitted curve for a sea level velocity of 1050ft/min. Then, at 14:42:46z and 74,889ft, the actual descent curve departs from the 1050ft/min curve and quickly, within one minute, joins the 780ft/min curve. This strange activity was caused by one of the payloads separating which left the parachute with a lighter load that slowed actual descent velocity. From this annotated set of curves we now knew when and where separation occurred: between 14:42:46z (74,889ft) and 14:43:46z (72,513ft) both at 41°09.78’N and 95°41.95’W. (From the recovered video we later discovered that separation was at 14:43:24z camcorder time that was known to be within a few seconds of GPS time.)

Now the task was to estimate a descent velocity curve for the separated free falling payload. The balloon tracking descent calculator again came into play by estimating the parachute diameter equivalent of the falling payload. The payload enclosure had dimensions of 10"H x 12"W x 7"D so I established bounds on equivalent parachute size as calculated from the average face area, 10.8in diameter, and absolute maximum, 12in diameter. The descent calculator was then used with the known separated payload weight to get an estimated range of possible sea level descent velocities, this one predicting 5248.7ft/min (60mph) for a 12-inch parachute.


Now we had everything needed for the first estimate of crash-landing coordinates. We had the altitude and coordinates of separation (used as "burst" altitude in WBALTRAK) and we had an estimated sea level descent velocity. These values were used to set up WBALTRAK for use with winds aloft from the morning RAOB flight from Omaha/Valley (OAX). Tracking output from WBALTRAK includes "burst" coordinates and descending coordinates at various time intervals.




 Note that right after separation, the free falling payload was descending at 23,634ft/min (268mph). WBALTRAK output predicted the time interval between "burst" (separation) and crash-landing to be 7 minutes. It also predicted drift in latitude and longitude increments that could then be applied to known separation coordinates to calculate a first approximation of where the payload crash-landed, 41°09.42’N 95°39.13’W (Prediction 1 on this map lifted from Mark's web page.)



This first approximation was explained to Mark and he immediately offered further information and a suggestion. The remaining payload and parachute landed further down wind that originally predicted. We compared predictions with what actually happened and found actual descent was along a track that was roughly 25% longer than Mark's preflight estimate. We extended the free falling payload track by the same 25%. The result was our second estimate of the crash-landing location, 41°09.443’N 95°38.576’W (Prediction 2 on the same map,)

Mark searched the area for several hours in at least three sessions but without the luck necessary to find a very small payload in a very big agricultural area. As he described the terrain, Mark could have been ten feet from the payload and not discovered it. An aerial image of the area suggests the magnitude of search effort that would have been required if a local rural resident had not discovered the payload.



Mark was able to salvage the video tape from the damaged camcorder. The tape has given us exact times of several important events:

 

WØRPK Home Page