Predicting Marine Traffic in the Ice-Covered Baltic Sea

Icebreaking activity and seasonal ice propose challenges for marine traffic prediction in the Baltic Sea. Traffic prediction is a vital part in the planning of icebreaking activities, but it remains largely as a manual task. The aim of this thesis is to examine factors influencing marine traffic modelling in ice-covered waters and propose a novel A*-based method for modelling traffic in ice. The current state of the marine traffic modelling and factors affecting vessel movement are concluded by examining the literature and historical vessel tracks. The field of traffic modelling research is growing rapidly. Currently the biggest challenges are evaluation of results and the lack of publicly available datasets. Moreover, the current approaches to model vessel movement in ice are promising but fail to capture how icebreaking activity influences vessel routes. The proposed model consists of sea, maneuverability, route and speed modelling. The model uses historical AIS data, topography of the sea, vessel type and dirways as main data inputs. The model is trained with summer tracks and dirways are used for modelling the ice channels kept open by icebreakers. The accuracy of the model is evaluated by examining route, speed, traffic and ETA (estimated time of arrival) prediction results separately. Moreover, the area between the actual and predicted route is introduced as an accuracy measure for route prediction. The model shows that winter route prediction can be improved by incorporating dirways to the modelling. However, the use of dirways did not affect the speed, traffic or ETA prediction accuracy. Finally, the datasets and source code used in this thesis are published online