Tsunami detection strategies and algorithms for ocean bottom pressure gauges in an early warning context

The last two decades have witnessed a rapid increase of awareness regarding tsunami hazard, especially since the 2004 Sumatra event. This has been accompanied by the development of Tsunami Early Warning (TEW) systems all around the world, which need instrumental monitoring networks to work. Classically, TEW systems have been based on seismic monitoring, although direct observation of travelling tsunami waves has acquired an ever more important role. The de facto standard tsunami measurement device today is the Ocean Bottom Pressure Gauge (OBPG), which measures the water pressure at the bottom of the sea. In this thesis, we test four tsunami detection algorithms, namely Mofjeld's algorithm, detiding with Empirical Orthogonal Functions, the Tsunami Detection Algorithm and a method based on the Fast Iterative Filtering (FIF) and IMFogram algorithms. FIF and IMFogram are data driven techniques for the decomposition and time-frequency representation of signals. We show that these techniques can be a compelling alternative to classical analysis methods, giving equivalent results, with the added robustness of data driven methods and the ability of performing multiple operations (denoising, tide removal, bandpass filtering) at once. Then, we leverage on their properties to develop a new tsunami detection algorithm. The four methods are tested against two datasets built from OBPG data from the NOAA's Deep-ocean Assessment and Reporting of Tsunamis (DART). The first dataset consists of month-long signals including only tides and noise, while the second includes day-long records acquired during past tsunami events. We discuss criteria to choose the optimal amplitude detection threshold, based on detection rates of tsunamis and earthquakes and false detections. The newly developed FIF-based technique shows promising results, both in terms of false detection rates and optimal detection thresholds. Finally, the ability of the technique to characterize the tsunami waveshape in real time are discussed in operational contexts are presented.