Browsing by Subject "bayesilainen mallinnus"

The interactions within plankton communities are complex, and realistic modelling of these interactions create a challenge in large-scale environmental models. The objective of this thesis was to evaluate whether Bayesian networks could be a suitable method in the modelling of these communities. Besides observing the interactions between different groups within phyto- and zooplankton communities, another goal was to focus on the potential change on the ecosystem level. To achieve this, dynamic Bayesian networks with hidden variables were used to observe whether structural changes in plankton communities could reveal larger trends in the aquatic ecosystem. To compare performance and accuracy of the model, two Bayesian food webs with differing causal links between observations were built. Of the two models, the simpler construct utilizing hidden Markov model fared better, and a clear trend was detected in the hidden variable. This trend in the time series signify that the relationships between the observed variables have changed during the study period. The plankton data set was collected from the Archipelago Sea between 1991 and 2016 and the results from the model were further analyzed alongside with this observational plankton data. In the samples the total biomass of phytoplankton grew throughout the study period, whereas at the same time the total biomass of zooplankton declined. As the Bayesian network considers the observable variables while maximizing the fit of the hidden variable, the observed trend in the hidden variable indicate that some unobservable variables are affecting both phyto- and zooplankton communities. This clear trend detected by the hidden variable might be related to a trend of increasing eutrophication in the study area, but to better understand the drivers causing this change further research is needed. Besides detecting underlying trends, the dynamic Bayesian networks are a promising method to study the interactions within plankton communities.