Data mining of spherical harmonic (SH) coefficients using artificial neural networks (ANN)

Gravity field recovery using space technology has evolved in the last two decades. Several dedicated satellite missions have been sent to the space to get more accurate and up-to-date gravity field information, including, Challenging Minisatellite Payload (CHAMP), Gravity Recovery and Climate Experiment (GRACE) and Gravity field and Ocean Circulation Explorer (GOCE), launched on 15 July 2000, 17 March 2002 and 17 March 2009, respectively. GRACE is the extended version of the CHAMP. CHAMP is an example of high-low satellite-to-satellite tracking (HL-SST) while the GRACE is an example of low-low satellite-to-satellite tracking (LL-SST) system. We study the gravity field in the form of SH coefficients using the range-rates observations from GRACE tandem satellites system. Sets of coefficients along with their standard deviation, recovered by GFZ-Germany up to degree and order 90 are available through the podaac data servers. In one month period, gravity field varies in few regions of the Earth. We observe it through the varying numerical values of few SH coefficients. In this contribution, we classify SH coefficients on the base of their information contents using artificial neural network (ANN) into two classes, one of them is the essential coefficient class which represents the varying gravity field and the other is the static coefficient class which does not have the varying gravity information. In the end we show that we can concentrate only on essential coefficients during the recovery process, rather than processing the whole set of coefficients.