Modelling phytoplankton populations in the Baltic Sea using spatial functional linear models

Specialeforsvar: Hans Vinther-Larsen

Titel: Modelling phytoplankton populations in the Baltic Sea using spatial functional linear models

Abstract: In this project I study phytoplankton measurements in the Baltic Sea from four national monitoring programs with measurements ranging from the late 1970′ies to the early 2020′ies. The phytoplankton populations are modelled as by station yearly mean curves and regressed against salinity and temperature measurement in function on function regression models. The samples are modelled as spatially dependent using tracevariogram models. The non-convex shape of the Baltic Sea is accounted for using a water distance metric derived from a delauney triangulation of the water body. Comparing the models using AIC scores, the spatial models are preferred to their non-spatial counterparts, though the variances become so large, that effects of covariates become hard to identify. Marginal 𝑅2 curves indicate that the random intercepts which accounts for the spatial dependence seem to compensate for wild behaviour in the driftfunctions leading, which in some cases seem to perform worse than the mean. Functional (spatial) models of cyanobacteria populations are seen to agree with a standard multivariate counterpart in the form of a generalized additive model as they are seen to prefer warm waters and low salinity. Dinoflaggelates are seen to show a non-concurrent effect of temperature, perhaps indicating a response to other biological functional groups, or some other unmeasured confounding variable. The spatial functional models in their current form do not seem to be the best fit for the data. In several places there is suspected to be a problem with identifiability, as the vast amounts of data has been aggregated to just 36 functional variables, which in particular in the variogram models is suspected to cause issues, as parameters in several fits take on extreme values.

Vejledere: Helle Sørensen, Jacob Carstensen, Aarhus Univerisitet
Censor: Ege Rubak; Aalborg Universitet