Evaluation of high rate real time GPS based tsunami warning system

Jet Propulsion Laboratory (JPL) is transmitting real time stream of GPS measurement over Internet via a UDP socket. It is a high rate stream that sends 1 Hz data from several stations, today the total number is 70 – 80 stations. This thesis is focusing on finding problems not only with the stream itself, also current problems with the actual measurement. There will be some improvement described however no focus is in network setup or geodetic detection of an actual earthquake. The delay of the stream is gradually increasing from 5 to 15 seconds and drops down to 5 seconds again. This delay is end to end time and have a rate of about one to one and a half hour. Over this time period JPL receives data, process it and resend it to end user. At the receiver end not all data arrives. There are 10 missing measurement in a row at the time the delay drops. The statement is that anywhere in the processing, either at JPL or at receiving computer there is a overflow in a buffer which explains the drop. There are other drops of data, which can be from seconds to hours in length. Whenever the dropped station starts to transmit again, sigma tends to be very large, which is a good thing. However it returns very quickly to normal, much faster than the actual GPS position. A sigma can take in the order of minutes to return while positions takes hours. This problem can be misleading and can create further problem for filters or other processing jobs where sigma is used. Different stations seem to be affected by the missing measure differently. This can explain the variation in result of displacement error. The measurement is subject to a centimeter to decimeter displacement error for a ten- minute gap of data. The displacement error is larger when ever a loss of data occurs. During the next coming hours a large variation of the measurement occurs and a displacement error will be much greater. The solution for this is either to minimize any drop of data which can be a matter of investing in better connection for the receiver stations around the world to JPL or look into the models JPL have and give a better sigma estimate. There are a significant autocorrelation in the signal, the peak can be up to 80% of top value and is close to a sidereal day. Because of this continues correlation a filter can be used to improve the time series. A five-day median filter gives an improvement up to 60%. This number will increase whenever the stated problems above are resolved.