Item

Abstract

Since launched in 2002, the Gravity Recovery and Climate Experiment (GRACE) has been proven to be a unique way to monitor total water storage (TWS) variations at large spatial scales (>300 km) by measuring Earth gravity changes and provides valuable information for hydrological and hydro-meteorological applications. In this thesis, globally gridded monthly-mean TWS variations are estimated by applying the state-of-the-art post-processing procedure, which has been evaluated first through a closed-loop environment by means of simulated GRACE-type gravity field time-series. In particular, the median scaling factors calculated from an ensemble of multiple global land model simulations do make the re-scaling more robust against particular weaknesses of a single model. The method to estimate gridded fields of measurement, leakage, and re-scaling errors, which can be used for further estimation of the basin-averaged TWS uncertainties, is also introduced. The TWS variations and error estimates are then applied to assess the accuracy of four global numerical model realizations and to identify the advantages and deficiencies of a certain model. Based on four different validation metrics, it is demonstrated that for the 31 largest discharge basins worldwide all model runs agree with the observations to a very limited degree only, together with large spreads among the models themselves. As a common atmospheric forcing data-set is applied to all hydrological models, it is concluded that those discrepancies are not entirely related to uncertainties in meteorologic input, but instead to the model structure and parametrization, and in particular to the representation of individual storage components with different spatial characteristics in each of the models. TWS as monitored by the GRACE mission is sensitive to the different model physics in individual basins and it could offer helpful insight to modellers for the future improvement of large-scale numerical models of the global terrestrial water cycle. In addition, the TWS variations and error estimates are also applied to assess skill scores of three different ensemble sets of decadal hindcasts performed with the coupled climate model MPI-ESM. Moderately positive skill scores of the initialized hindcasts are obtained both with respect to the zero anomaly forecast and the uninitialized projections in particular for lead year 1 in moderate to high latitudes of the Northern Hemisphere. Changes in the initialization and increased resolution implemented in the different experiments indeed lead to more skillful initialized hindcasts than in the earlier experiments, thereby documenting improvements of the MPI-ESM decadal climate prediction system during the most recent years.