Efficiency of different signal processing methods to isolate signature characteristics in altimetric water level measurements

Observed time-series of water transport in rivers can be perceived mathematically as a superposition of non-linear long-term trends, periodic variations, episodic events, colored instrument noise, and other components. Various statistical methods are readily available to extract and quantify both stationary and non-stationary components in order to subsequently attribute parts of the signal to underlying causal mechanisms. However, the available algorithms differ vastly in terms of computational complexity and implicit assumptions, and may thus have their own individual advantages and disadvantages. By employing a suite of time-series analysis methods for 1D (Wavelets, Singular Spectrum Analysis, Empirical Mode Decomposition) and additional statistical assessments like Pruned Exact Linear Time (PELT) tests for change point detection, we will analyze data from two virtual stations at Elbe River (Germany) and Urmia Lake (Iran) that are representative for the central European region with a rather humid climate, and the more arid conditions of Central Asia with much smaller hydrological signal variations, respectively. It is in particular the latter region with a much less developed in situ hydrometeorological observing system, where we expect that carefully processed geodetic data might contribute most to the monitoring of large-scale terrestrial water dynamics. This contribution will highlight the benefits of more advanced signal analysis methods for extracting relevant hydrometeorological information over more conventionally applied algorithms.