Sélectionnez votre langue
Sélectionnez votre langue
Machine learning methods for processing big Earth data Geomorphological landforms of the oceanic trenches, their formation and variation of the geometric shapes is a question of special importance to the scientific community in marine geology. The actuality of this question has significantly increased since the beginning of the rapid development of the IT tools and methods of the advanced data analysis, yet its understanding remains patchy. Since the majority of the oceanic trenches are located along the margins of the Pacific Ocean, it plays a central role for their analysis and understanding their formation oceanic trenches. Specific geological conditions, presence of the tectonic subduction zones, vast territory of the Pacific Ocean with complex circulation system, extension of the ’Ring of Fire’, a seismically active belt of the earthquakes and volcanic, make the trenches of the Pacific Ocean highly sensitive to the factors affecting their formation which cause variations in their geomorphic shape forms. In this context, the most representative indicators of the variations in the deep-sea trenches are geological and tectonic factors, such as dynamics of lithosphere crust affecting speed and intensity of plates subduction, magnitude and frequency of the submarine volcanoes causing active sedimentation. Nowadays, studying marine geological phenomena and complex processes by programming and scripting has been a powerful method. Rapid development of the advanced methods of data analysis presented such effective tools as GMT, Octave/MATLAB, R and Python. It is particularly efficient when applied to the massive amounts of marine geological data. Big data processing by advanced scripting is a crucial approach, as algorithms of libraries give access to the accurate and rapid data analysis [373]. Specific information about distant and hard-to-reach deep-sea trenches can be gained for precise visual- ization and analysis of their submarine geomorphology from local to regional and global scales. However, despite all the efforts, there is a lack of uniformity in studying deep-sea trenches, a shortage of systematic mapping of the Pacific trenches and a lack of understanding of the geomorphological variation between the trench profiles in different parts of the ocean: southern and northern, eastern and western, and their response to the geological and tectonic local settings in the places of formation. Therefore, this dissertation develops a systematic approach to monitoring and comparative analysis of the geomorphological shape forms of the deep-sea trenches formed under specific geological and tectonic conditions along the margins of the Pacific Ocean. The study area encompasses Pacific Ocean, and more specifically, includes 20 selected target trenches: Aleutian, Mariana, Philippine, Kuril-Kamchatka, Middle America, Peru-Chile, Palau, Japan, Kermadec, Tonga, Izu-Bonin, New Britain, San Cristobal, Manila, Yap, New Hebrides, Puysegur, Hikurangi, Vityaz and Ryukyu. These are the major trenches of the Pacific Ocean and, therefore, the most representative for the geomorphological modelling. This dissertation identifies tectonic plates formation, slab subduction, historical geological development, earthquakes and submarine volcanoes as the primary types of impact factors affecting trenches formation. Secondary factors include ocean currents, sedimentation and biota contributing to the sedimentation. Seafloor geomorphology in hadal trenches is strongly affected by a variety of factors that necessarily affect the shape of their landforms. Using data modelling, the shapes of the profiles transecting the trenches in an orthogonal direction were compared and analyzed in order to highlight the differences and variations in their geomorphology. The objective of this PhD study is to perform a geomorphological classification of the shape forms of the trenches through ordering them into groups base don the common characteristics of the trenches’ landforms in plan and attaching labels to these groups. Following geomorphological profile shape types have been identifies and trenches are classified into seven types: U-formed (in plan), V-formed (in plan), asymmetric, crescent-formed, sinuous-formed, elongated, cascade-formed. For each type (U, V, asymmetric and so on) characteristic steepness sub-types are identified: strong, very strong, extreme, steep, very steep. Valley slopes are classified as follows: very high, high, moderate, low, based on the curvature degree. Size and valley slope classes are analyzed in the context of physical environment and tectonic and geological development of the area of trench formation.Technical aim of this PhD study was to experiment with and extend current methods of geospatial modelling for geomorphological classification of the submarine landforms of the trenches. Using methods of the advanced data analysis is crucial for the precise and reliable data processing, since understanding seafloor landforms can only be based on the computer-based data modelling due to their unreachable location. The selection of the methodology, tools and algorithms is explained by research objectives and goals. The specifics of the marine geology consists in the high requirements towards data processing. Datasets were processed, computed and analyzed in semi-automatic regime by Machine Learning (ML) approaches, using advanced algorithms of data analysis and effective visualization through application of the advanced programming tools and Generic Mapping Tools (GMT) scripting toolsets.This dissertation presents an automated workflow enabling large-scale profile cross-sectioning aimed at transect geomorphological mapping, quantitative comparative analysis and classification of the 20 trenches of the Pacific Ocean. The methodology of the GMT includes algorithms of sequential scripting for the cartographic visualization and mapping, automatic digitizing of the cross-section transect profiles, and geomorphic modelling of the trenches. In total 50 modules of GMT scripting toolset were trained on extensive datasets collected from 20 trenches across the whole region of the Pacific Ocean. Using high-resolution bathymetric datasets (General Bathymetric Chart of the Oceans (GEBCO), ETOPO1 and Shuttle Radar Topographic Mission (SRTM)), sample transects of the trenches were modeled, analyzed and compared. Variations in shape forms, steepness and curvature were analyzed by computed models for each trench. The tables were converted from QGIS plugins to Python libraries and R packages, and from GMT to Octave via AWK languages. The results revealed variation in the shape and steepness of the submarine geomorphology in 20 trenches of the Pacific Ocean. A strong correlation between the geomorphic profile shapes with geological factors and level of tectonic activities (earthquakes, volcanism, speed of tectonic plate subduction) and the scale of trench steepness, curvature and shape unevenness is confirmed and analyzed. Geomorphological structure of the trenches and dynamics of the tectonic plates subduction are analyzed and assessed at each trench regionally (north, south, west and east Pacific). The novelty of the study consists in presented systematic classification and comparative modelling of the geomorphic profiles of the deep-sea trenches by means of the sequential usage of the advanced scripting toolsets. Technical innovativeness consists in a combination of GIS, GMT, Python, AWK, R.The actuality of this dissertation lies in its strongly multi-disciplinary nature demonstrating a com- bination of the following approaches: 1) systematic multi-source geospatial data analysis; 2) statistical data modelling and processing by libraries of the Python and R, AWK and Octave/Matlab; 3) geological literature analysis; 4) cartographic mapping and modelling by GMT shell scripts and visualization in QGIS. Technical scripts used for advanced statistical analysis are presented in full in the Appendix A for future replication and reproducible analysis in other trenches of the World Ocean.