Pinpointing A Crash Site: San Diego Analyst Shares Lessons From Air France Search
April 10, 2014 12:56 p.m.
Colleen Keller, senior analyst, Metron Inc.,
MAUREEN CAVANAUGH: This is KPBS Midday Edition, I am Maureen Cavanaugh. An Australian search plane is the latest to pick up a signal that may be from the lost Malaysian flight 370. The world has followed the search for the missing passenger jet for over a month now, the airliner and 239 people on board were last heard from on March 8 on a flight from Kuala Lumpur to Beijing. Officials say the search area is now the smallest of this ever been that it is still a sizable area in the Indian Ocean. My next guest knows a little bit about pinpointing the location of a missing plane. Colleen Keller is a senior analyst with Metron Incorporated in San Diego, she was on the team responsible for locating Air France flight 447 after it crashed into the Atlantic Ocean. The method she used was not technology but mathematics. Welcome to the program. What kind of data going to calculations when you're trying to come up with the probabilities for a search?
COLLEEN KELLER: The approach that we use is to consider all of the hypotheses for what happened to the aircraft or whatever we are searching for. That can include for the case of the Malaysian aircraft, hijacking, terrorism, or malfunctions. Each of these theories have data that support them and refute them. Some of that data can quantify the position. For instance, UF radar information that is usually very valuable because it gives you an exact location and is pretty precise and gives you a multitude and speed. And the radar indicated that it was turning back towards the south, towards where it came from or maybe into the Indian Ocean. That does not support the hypothesis that hijacking was hypothesis because it's not heading toward a country that harbors known terrorists. It indicates that there is perhaps malfunction or a suicidal intent or hijacking gone awry, but the aircraft was no longer in control.
MAUREEN CAVANAUGH: You put this information into an actual mathematical system?
COLLEEN KELLER: It is a way to combine different hypotheses altogether into a single representation of what you know. At any one time you can look at a map and say all of the evidence is supporting airplanes in this area but we still have a likelihood that it might be in the northern areas or in the Gulf of Thailand etc.
MAUREEN CAVANAUGH: When you have combined data that you have, what kind of guidance can you give? Do you give them various scenarios and they choose or to give them this is the most likely place?
COLLEEN KELLER: We work with searchers to arrive at the hypotheses and the hypothesis of maintenance very other is failure is very wavered but favored, but it could be a hijacking. We take input from all of the experts from radar and the satellite pings and the endurance on the aircraft from Boeing and we factor all of that in, but the map that we are giving them is actually a map of the search area of the greatest possible area it could be in. All gritted out, into maybe a 10 x 10 grid, and in each grade be assigned probability that the aircraft is there based on all of the information that we know and that is what we call a probability map. It is using these DRAM to incorporate all of these sources of data and their uncertainties, and then sum them all up into what we call a priority distribution before any searching has happened of where the airplane likely hit.
MAUREEN CAVANAUGH: That is remarkable, the Air France flight that you're involved in trying to track, that took almost 2 years to locate, what where the difficulties you encountered there?
COLLEEN KELLER: That was an interesting case, we generated a prior distribution based on historical loss of control accidents of other aircraft and aerodynamic modeling of where the aircraft could of gone from. To the surface, how far forward and how quickly would fall, and we also took drifting wreckage and also drifted it back to where it would likely hit the surface and we combined all of those things and came up with a prior distribution the next step using Bayes Theorem he is to evaluate the effort that they used to find stuff, like surface searches looking for debris.
MAUREEN CAVANAUGH: Are you constantly adding data to this?
COLLEEN KELLER: Exactly, Bayes Theorem allows you to update your search of knowledge with every search you conduct that does not produce a find and we fold in that negative information and we do the probability map with new data, is always sewing showing progress through the search.
MAUREEN CAVANAUGH: In that particular instance, once you learn that both of the black boxes in the aircraft disaster malfunction, that completely redid the probability data and this search location that you said that the aircraft will most likely be in.
COLLEEN KELLER: They didn't he can search just like they are doing now for the Malaysian aircraft, they did an exhaustive search along the aircraft intended track and they cleared that area. Clearing some say you are 100% sure [ CHECK AUDIO ], but in this case they were looking for something that was not there because both of the beacons on both the black boxes had been damaged, and they were no longer functioning with the aircraft hit the water or after it hit the water. So, the search for this signal was ineffective, there was no signal and what we did after two years, they did that search and then they started looking elsewhere because then be given very high probability of having detected something if it was there, they look elsewhere and they looked in all of these different other places and then they asked us to update the map one last time and we consider that perhaps the beacons were both broken. We reflected that into the map and all the sudden the initial search area came up and stuck out like a sore thumb.
MAUREEN CAVANAUGH: You seem pretty sure now that those pings are there and those are man-made sounds, and they are thinking that they are coming from this plane, is that therefore that they don't need any more probability data? If they know now where the most likely places that these planes went down?
COLLEEN KELLER: We can still keep track of it using Bayes Theorem but it's such a small search area at this point, that they can physically search the entire area, but it would take some time and money. But they can send a submersible down with sites canst not sonar and map an image the whole area looking for the debris field and it may be difficult because it is an tough rocky terrain, and the landing gear and the engines won't stand out, they may actually have to do it by camera which even gets to more like the needle in the haystack kind of thing but they are in a workable problem now, they really needed this before when they didn't know what ocean they're looking at.
MAUREEN CAVANAUGH: Exactly, north-south, the whole curve there on for a while. I think the overarching thing about all of this is that people are at fault that in this day of age that something as big of an airliner can vanish but you have been involved in at least one search for a plane that did vanish, does that continue to surprise you as well?
COLLEEN KELLER: Now, my hobby when I'm not at work is searching for missing airport airplanes, I am picking of cold cases and looking for aircraft's. They're not all big aircraft, they are often small airplanes that don't have satellite systems, it's a big world out there and searches in the inefficient, and people do not realize they can looking twice in one place and still miss something and it is just a lot of open space out there.
MAUREEN CAVANAUGH: All right, I guess we forget about that. We feel that we are being tracked all of the time wherever we are.
COLLEEN KELLER: Not in the Indian Ocean.
MAUREEN CAVANAUGH: Do believe that will find a Malaysian fight 370?
COLLEEN KELLER: I believe that they were not going to find it before but I hate to go on record being conclusive but it is looking very good, these beings are exactly what we needed and it was a minor vehicle that they found the pings to begin with, is the detection range of that sensor and they are very fortunate to have some things. I think now that we're on to them from the black boxes, it's just a matter of time if they want to put the money and time into it they can find it.
MAUREEN CAVANAUGH: And timeframe? Would you think?
COLLEEN KELLER: I think they might put the bluefin twenty-one submersible in the next week to get as much data out of those things as they can because it's an opportunity they don't want to miss and you can't run the bluefin along side the listing device. With the bluefin goes down can put it right down on top of the wreckage or it can take months but I think it were a cup couple of weeks away from a solution.
MAUREEN CAVANAUGH: I have been speaking with Colleen Keller, thank you very much.