Deep-sea polymetallic nodules spatial modelling with machine learning algorithms and benthic plume monitoring
This dissertation studies a) the spatial distribution of deep-sea polymetallic nodules using autonomous underwater vehicle data, including multibeam echosounder data, seafloor images, and photomosaics, which provide a m-to-mm spatial resolution that enables detailed geomorphological analysis and image-derived estimations of polymetallic nodule coverage. Multi-model machine learning approaches, including generalised linear models, generalised additive models, random forests, support vector machines, and neural networks, are used to link the different datasets. They predict well (R² > 0.7) for geographical areas where correlations between the response variable and predictors exist, the spatial autocorrelation is isolated from the model training, and the training data ensure an adequate representation of the feature space. Quasi-linear, monotonic, and nonlinear relationships between nodule coverage and spatial predictors were identified using interpretable machine learning tools, such as variable importance and partial dependence plots. b) the deep-sea mining-induced benthic plumes, showing that they initially spread as a high-density turbidity current, with little to no mixing with the ambient water. The turbidity current is channelled through steeper seafloor sections (> 3°), increasing the propagation distance outside the mining site. After the weakening of the turbidity current dynamics, a passive transfer of suspended particles by bottom currents continues to transport fine-grained particles. The suspended particulate matter concentration (mg L-1) decreases rapidly with distance and altitude, reaching almost at background levels at the end of the monitored area, 4.5 km away. Particle flocculation results in sediment redeposition over short distances (< 500 m) from the mining site. Within the mining site, a mm-resolution seafloor photomosaic reveals sediment redeposition of 3 cm and seafloor erosion of a min. 5 cm.
Vorschau
Rechte
Nutzung und Vervielfältigung:
Bitte beachten Sie, dass einzelne Bestandteile der Publikation anderweitigen Lizenz- bzw. urheberrechtlichen Bedingungen unterliegen können.