Water bodies such as rivers, estuaries, ports and coastal waters are usually monitored at a limited number of locations and moments in time. These can be fixed monitoring stations, buoys, ship-based sampling campaigns or occasional manual measurements. Parameters such as temperature, salinity, dissolved oxygen, turbidity, pH or nutrient concentrations nevertheless vary continuously in space and time. This thesis investigates how a complete and reliable picture of that variation can be reconstructed from sparse point measurements, and how much confidence can be placed in the result.
The student is expected to start with a literature review of the relevant methods. These range from classical spatial interpolation techniques such as inverse distance weighting and kriging, through statistical and regression-based approaches, to data-driven methods such as Gaussian process regression or machine learning models. Where relevant, the review should also cover how these methods can be combined with physical knowledge of the water system, for example tidal dynamics, river discharge or mixing processes. The review should end with a reasoned choice of one or more methods to apply.
The core of the work is practical. The student will select and prepare a suitable dataset of point measurements, for example from a public monitoring network or from measurement campaigns made available by the supervisor. They will explore its structure, gaps and quality, and then implement the chosen modelling approach, preferably in Python or R. The model's performance must be evaluated rigorously, for instance through cross-validation or by withholding stations. Particular attention should go to quantifying the uncertainty of the estimates. A key question is how the density and placement of the measurement points affect the quality of the reconstruction.