W ramach Seminarium ZGP referat pod tytułem: Temporal correlations in GRACE time-series wygłosi: Pani M.Sc. Ozge Gunes,
doktorantka z Turcji, odbywająca staż naukowy w Wojskowe Akademii
Technicznej pod kierunkiem Pani dr hab. inż. Anny Kłos
Autorzy referatu: Ozge Gunes and Cuneyt Aydin, Yildiz Technical University, Faculty of Civil Engineering, Istanbul, Turkey
Seminarium odbędzie się 8 kwietnia 2022 (piątek), o godz. 13:15,
via webex.
Streszczenie: GRACE (Gravity Recovery and Climate Experiment) has been providing data on the Earth’s gravity field since 2002, which has proven to be particularly sensitive to hydrological events and their changes over time. It is used in a variety of geodetic and geodynamic research, including monitoring changes in total water storage, mass loss in glacial regions, sea level changes due to glacier melting, variations in the amount of water in the oceans and bottom pressure. The equivalent water thickness (EWT) geopotential function is used in GRACE-related studies to analyze hydrological events on the land, basin by basin or mascon by mascon. The EWT can be derived from GRACE satellite data (Level 1), monthly spherical harmonic models provided by some process centers (GFZ, CSRS, etc.), and time-series of points or mascons on the physical surface (Level-3 data) published by GSFC, CSR and JPL. A trend, annual and semi-annual sinusoidal signals are components of an EWT signal for each data type. In order to estimate the trend of each point or mascon, these periodic signals are used. However, for time series analysis, simply applying the regression model may not be sufficient. Because geodetic time series may have temporal correlations, colored noise in the data can be caused by such correlations. Besides white noise, our research shows that the GRACE time-series also has colored noise. These correlations are well known in geodetic modeling studies, and they can lead to misinterpretations of parameter estimates like the rate of change or the amplitudes of signals if they aren’t considered. In spectrum analysis, a variety of methods can be used to deduce the presence of these noises. In time-series analysis, one of the procedures used is the analysis of power spectral density (PSD) functions. The PSD log-log plots of the data (or the residuals) provide information about the noise’s characteristics. Implementing the noise model of white noise+flicker noise, white noise +random walk, etc., maximum likelihood and variance component estimation methods can be used to analyze these correlations and describe the colored noise in the GRACE time series. These methods are good at estimating the amplitude of each noise in the assumed model for a GRACE time-series, but they may not work in all cases. Instead of using geodetic noise modeling methods, autoregressive models such as ARIMA may be used because they only require specifying the relationships between observations at specific lags in the time-series, resulting in the residuals becoming stationary as a result of the analysis. Thus, in this study the autoregressive noise modeling with the ARMA/ARIMA model is considered to obtained realistic estimates in the GRACE EWT time-series. For this purpose, GSFC Mascon solutions have been analyzed. Preliminary results are discussed in terms of trend estimation and significance, forecasting, and compatibility with the GRACE-FO mission.
