Global ve bölgesel olarak GRACE L2 verilerinin PCA ile analizi

The satellite gravity mission Gravity Recovery and Climate Experiment (GRACE) provides a huge amount of data which allows to quantify temporal and spatial variations of the Earth's gravity field caused by mass transport and mass distribution. However generating and using GRACE-derived estimates of total water storage (TWS) for validation purposes is a complicated task due to the nature of the GRACE Level 2 (L2) products and their spatio-temporal resolution limitations. This results in flawed data-sets. For instance, the GRACE L2 results contain stripe shaped correlated noise, artifacts oriented from north to south, and ring shaped ones about centers of large masses. Besides, spherical harmonic coefficients are dominated by noise at high degrees. Therefore, spherical harmonic expansion has to be cut-off at an optimum maximum degree/order with keeping signal-loss as little as possible. This truncation of the input signal causes amplitude dampening and signal leakage outside of the area of interest. Also missing months of GRACE as a result of battery management causes gaps in the time-series which affect results of the analyzes. Within this study, at first post-processing steps of GRACE data are studied to have optimum condition of our data-set. The contaminating effect of the noise is demonstrated and filtering techniques as a solution for the correlated noise are introduced. Then the performance of the Gauss filter to remove noise is assessed. Gaussian filtering removes random noise, but leaves the correlated noise intact. Following that, truncation and leakage problems for Amazon, Greenland, Mekong and Tigris/Euphrates regions are observed. A test data set, obtained by a simulation of continental hydrology, and GRACE-like example data derived from it, both supplied by Helmholtz-Zentrum Potsdam-Deutsches GeoForschungsZentrum (GFZ), are used to test truncation in the spectral domain. Since the GRACE-like data is acquired from simulated continental hydrology data, the latter serves as a reference data (true signal). Because that way the true signal is available, the GRACE-like data can be used to recover the restricted spatial resolution of GRACE data and allows making the analysis in the spectral domain without any signal loss. For this analysis the spherical harmonic synthesis in equivalent water height up to 50 and 90 degree and order is executed in the spectral domain. The results are compared with the reference signal for the regions Mekong, Amazon, Greenland, Turkey and Tigris/Euphrates. Besides, masks that minimize signal-loss are enlarged for the regions Amazon, Greenland and Tigris/Euphrates. Enlarging the region mask covers more signal, however it can not be guaranteed, that the recovered signal belongs to the region of interest. As a second step, the statistical signal-signal decomposition technique Principle Component Analysis (PCA) is applied to the data-sets to reduce dimensionality and determine the direction of the highest variance of the data. PCA solutions of three data-sets (CS2, CS3 and CS6) obtained by applying different post-processing techniques are compared each other. In the first mode, the results showed identical trend in each region. The CS6 data-set resulted in smoother signal than the newer releases in all modes and all regions. The effect of gaps in GRACE time-series (due to battery management) is analyzed by adding artificial gaps to GRACE data between 2003-2010, which did not previously have any gaps. Comparing results yields that missing monthly data caused 1.54 % signal loss in Greenland, 0.71 % in Amazon and 0.22 % in Mekong. Apparently and to no surprise, the signal loss is bigger the larger the area is. The performance of the conventional Principle Component Analysis (PCA) is investigated by applying it to a monthly series of Gauss filtered (400 km radius) GRACE andWaterGAP Global HydrologyModel (WGHM) data for the years 2005 to 2008. The results are compared globally and for Amazon, Orinoco, Turkey and Tigris/Euphrates regionally. Besides the signal's reconstruction is accomplished for desired variances, 80 % and 95 %, globally and regionally. The required PCA mode numbers with respect to the regions of interest are determined. As it turns out, this depends on the region. For 95 % variance globally mode 8, in Amazon mode 3 and in Orinoco mode 1 are required. WGHM and GRACE data showed the same trend for each region, however the amplitude of WGHM was lower in every case. The PCA results of all 3 previously mentioned GRACE data-sets provided by GFZ are also validated with in-situ tide gauges over the Mekong basin between 1st of July and 30th of November for the years 2008, 2009, 2010 and 2011. The results depict that consistency of GRACE and in-situ data change depending on the corresponding year.