| description abstract | his study addresses in general terms the problem of the optimal combination of multiple observation datasets. Only satellite-retrieved geophysical parameter datasets are considered here (not the raw satellite observations). This study focuses on the terrestrial water cycle and presents methodologies to obtain a coherent dataset of four water cycle key components: precipitation, evapotranspiration, runoff, and terrestrial water storage. Various innovative ?integration? methodologies are introduced: simple weighting (SW), constrained linear (CL), optimal interpolation (OI), and neural networks (NN). The term ?integration? will be used here, not ?assimilation,? as no model will be included in the data fusion process. A simple postprocessing filtering (PF) step can be used to impose the water cycle budget closure after the integration method. It is shown that this constraint actually improves the estimation of the water cycle components. The integration techniques are tested using real observation data over the Mississippi and Niger basins from satellite and in situ measurements. A Monte Carlo experiment with a synthetic uncertainty perturbation model is used to measure the ability of the SW, OI, and NN, with or without the PF step, to retrieve the four water cycle components. Once the PF closure constraint is added, the methodologies have equivalent accuracies. The need for these types of methodologies should increase in the future since multiple observation datasets are now available and the climate community needs to combine them into a unique, optimal, and coherent dataset of multiple parameters. A companion paper will test these methodologies on satellite observation datasets at the basin and global scales. | |
