Machine learning and complex network techniques to inform marine biogeochemistry modelling and data assimilation

Higgs, I. (2025) Machine learning and complex network techniques to inform marine biogeochemistry modelling and data assimilation. PhD thesis, University of Reading. doi: 10.48683/1926.00130697

Abstract/Summary

This thesis advances the understanding and forecasting of shelf sea ecosystem dynamics by integrating complex network theory and machine learning into the modelling and data assimilation of marine biogeochemistry. Firstly, using a state of-the-art coupled physical–biogeochemical model of the North-West European Shelf (NWES), complex network analysis is employed to identify functional groups and spatial connectivity within the ecosystem, revealing the scale and structure of biogeochemical interactions. These networks show how information could prop agate through the system and help to delineate coherent regions of strong in ternal connectivity. Secondly, the thesis addresses limitations in current data assimilation (DA) techniques, which struggle with the seasonal complexity, lack of flow-dependence, and observational sparsity typical of marine biogeochemistry. Machine learning (ML) methods are introduced to enhance DA performance by statistically linking observed and unobserved variables. In a 1D prototype sys tem, ML-driven schemes demonstrate improved update accuracy for unobserved variables and exhibit promising transferability across spatial domains. Together, these approaches offer a pathway toward more efficient and scalable forecasting sys tems, with implications for future research in marine biogeochemical modelling, operational marine forecasting, and the data-driven understanding of shelf sea ecosystems.

Item Type	Thesis (PhD)
URI	https://centaur.reading.ac.uk/id/eprint/130697
Identification Number/DOI	10.48683/1926.00130697
Divisions	Science > School of Mathematical, Physical and Computational Sciences > Department of Meteorology
Download/View statistics	View download statistics for this item