Use of the autocorrelation function in EOF analysis of GRACE data

Abstract

The GRACE mission launched with a pair of satellites, orbiting in a near-polar orbit, at a distance of about 220 km from eachother, to map the time-variable earth’s gravity field, since 2002. The study of time-variable gravity field has been proved to be very helpful in climate science studies. The gravity variations in the GRACE observations are mass variations inside the earth, exchanges between glaciers and oceans, changes due to surface and deep currents in the ocean. Monthly maps are used to study these gravity variations. The raw gravity field data obtained, is too much noisy and the main source of this noise is north-south stripes which is due to polar orbit of satellites. Due to these noisy stripes, filtering of GRACE time-variable gravity field is required. In this thesis, Empirical Orthogonal Function (EOF) analysis is used to filter and analyse the GRACE gravity data. The method is used to extract the dominant variations by reducing the dimensionality of a dataset, among a group of time-series data. This dimensionality reduction and extraction of dominant variance is achieved by linear coordinate transformation to a new set of basis vectors via singular value decomposition. The decomposition gives spatial and temporal components along with variance values. Temporal components are analysed by the dominant variance rule, Kolmogorov-Smirnov rule and autocorrelation function respectively, in order to recover signal from noise. Dataset recovered by dominant variance rule, reduces the striping but it may remove the signals as well, especially the signals from ocean. The level of signal reduction is less in Kolmogorov-Smirnov rule, whereas autocorrelation performs well in comparison to both. Geophysical signal reduction is very less in using the autocorrelation function for filtered data analysis and the results are even much better if both Kolmogorov-Smirnov rule and autocorrelation results are combined together. Thus, autocorrelation can be a better approach to select the signal components from the noisy ones. EOF anlaysis is explained with its theoretical background and then its application on the GRACE data. The focus is on the use of autocorrelation function and its performance in the filtered data analysis. Here, the entire procedure is applied in spatial domain on the processed equivalent water height values. In future, autocorrelation function can be used for data analysis in spectral domain for more better results. Its performance can be evaluated on regional analysis basis.