To main content

Towards precision state estimation of the ocean using the Ensemble Kalman Filter

Abstract

The ocean is central to climate, ecosystems, food production, transport, and coastal industries, yet much of its variability remains difficult to observe directly. This limits our ability to describe present conditions, understand how the system evolves, and provide reliable information for forecasting and coastal decision-making. Variability unfolds acrossmultiple spatial and temporal scales, from basin circulation to coastal fronts and fjord exchange, and much of it occurs below the surface. No single observing system can resolve this full range. The observational pyramid provides one useful way to frame the problem, with satellites offering broad synoptic coverage, mobile and autonomous platforms resolving evolving regional features, and fixed or depth-resolving systems constraining local structure and continuity. Numerical ocean and ecosystem models provide the dynamical link between these observational levels, but they are also uncertain. This challenge is addressed here through data assimilation, where observations are combined with model predictions to estimate the evolving ocean state, and through observation-design methods that help determine where measurements are most informative. The Ensemble Kalman Filter is implemented in SINMOD, an ocean model developed and maintained at SINTEF Ocean, to combine heterogeneous observations with a coupled ocean and ecosystem model and to represent uncertainty through ensembles on high-performance computing clusters. The work shows how observations can constrain both observed and unobserved system states, including dynamical model variables and static parameters, and it identifies practical requirements for stable estimation such as localization, perturbation design, and treatment of bounded ecosystem variables. The same framework is then used to study how analysis corrections propagate from regional mother domains into nested coastal domains. The results show that this impact depends strongly on physical connectivity and is most persistent where corrected water masses enter through inflow. The framework is also applied in two observing contexts. A two-month operational field campaign demonstrates closed-loop monitoring in which chlorophylla observations from the HYPSO-1 satellite and the AutoNaut autonomous surface vehicle are assimilated into operational nested SINMOD forecasts, while those forecasts provide feedback to adaptive sampling under practical constraints related to latency, communication, weather, computing, and model bias. A complementary observation-design study in the Trondheim Fjord uses a large high-resolution ensemble to optimize platform type and placement by expected posterior uncertainty reduction within a region of interest, and compares the predicted impact patterns with an assimilation case based on the OceanLab observing system. Overall, the work links multiscale observing, coupled modelling, data assimilation, adaptive monitoring, and observing-system design within one coherent framework for coastal and fjord environments.

Category

Doctoral thesis

Language

English

Affiliation

  • SINTEF Ocean / Fisheries and New Biomarine Industry
  • Norwegian University of Science and Technology

Date

02.06.2026

Year

2026

Publisher

NTNU Norges teknisk-naturvitenskapelige universitet

Issue

2026:258

ISBN

9788235301604

View this publication at Norwegian Research Information Repository