Résumé
The scope of this report is to present the science developed within the VECTORS project to improve the understanding of the key processes driving the behaviour of human agents utilising a variety of EU maritime domains. While particular attention has been paid to the spatial interactions between fishing activities and other human uses (e.g., maritime traffic, offshore wind parks, aggregate extractions), the behaviour of non-fishing sectors of activity has also been considered. Various quantitative and semi-qualitative approaches have been pursued to gain better insight into behavioural drivers based on past data, and also forecast how human agents would react if access was constrained by either management (e.g. Marine Protected Areas – MPA), or the installation of a new operator. This report covers the North Sea and Eastern Channel, and also one area of the Baltic Sea: the Gdansk Bay. Fine-scale catch and effort data by fishing vessel, fishing trip, gear used and ICES rectangle visited have been made available for the French, Dutch, English and German fleets. The VECTORS WP2.3 team has also collected data from the non-fishing sectors of activity, in particular, aggregate extractions and maritime traffic. The objective of collecting data for non-fishing sectors is to produce a Spatial Overlap Metric measuring the constraint exerted by other sectors of activity on fishing. Comprehensive aggregate extraction and shipping intensity metrics could then be derived dynamically at a fine spatial and temporal resolution. For the other sectors potentially competing for space with fishing (e.g., wind farms, oil/gas extractions, aquaculture), and also for protected areas, a static overlap metric has been set to the surface occupied by the plant or area protected. In order to apply common methodologies and codes across different case studies, whilst abiding by confidentiality issues around these data, it has been decided to develop a common exchange format to collate the data used in subsequent analyses; five tables have then been produced. Two complementary types of approaches have been carried out to analyse and/or model the mechanisms of human behaviour, which are hereby referred to as quantitative and qualitative research. Quantitative research consisted of analysing fishing decision-making processes based on existing data and then making forecasts building on scenarios, while qualitative research consisted of interviewing stakeholders from different sectors of activity to get their views on both their past and likely future behaviour. Different methodological approaches have been pursued by different institutes, and these were applied to several case studies wherever possible. Modelling the current and past dynamics of fishing vessels: The understanding of the dynamics of fishing vessels is of great interest to define sustainable fishing strategies and to characterize the spatial distribution of the fishing effort. It is also a prerequisite to anticipate changes in fishermen’s strategy in reaction to management rules, the economic context or the evolution of exploited resources. In this context, analysis of individual vessel's trajectories offers promising perspectives to describe behaviour during fishing trips. A hidden Markov model with two behavioural states (steaming and fishing) was developed to infer the sequence of non-observed fishing vessel behaviour along the vessels' trajectory based on GPS records. Conditionally to the behaviour, vessels movements were modelled by a discrete time solution of a (continuous time) stochastic differential equation on vectorial speeds. The model's parameters and the sequence of hidden behavioural states were estimated using an Expectation-Maximization algorithm, coupled with the Viterbi algorithm that captures the most credible joint sequence of hidden states. A simulation approach was performed, that outlined the importance of contrast between the model’s parameters as well as the influence of path length to allow good estimation performances. The model was then fitted to four original GPS tracks recorded with a time step of 15 minutes derived from voluntary fishing vessels operating in the Channel within the IFREMER's RECOPESCA project. Results showed differences in parameter estimation depending on the gear used, on both the speeds during fishing operations and the Markovian transitions between behaviours. Results also suggested the benefits of future inclusion of variables such as tidal currents within the ecosystem approach of fisheries. Hidden Markov models are well suited to describe jointly fishing boat movement and associated fishing activities. They allow us to estimate the sequence of activities (i.e. fishing, travelling) along a trajectory, as well as the movement parameters (speed, turning angle) associated with each activity. Normally, these models are developed to characterize the spatial dynamics of fishing vessels that belong to a specific fishery with a given métier. However, because of the large variability that exists in fishing practices, some adaptations in the modelling structure are needed when the spatial dynamics of one or several fishing fleets present a mixture of métiers with distinct traits of movement and trajectory. A procedure was developed to capture the variability of fishing practices and associated vessel trajectories. Fishing trips were characterized by their métiers, which were identified for each gear by clustering landing profiles (in value). Fishing boat trajectories were described using movement parameters (speed, acceleration, turning angles, straightness) estimated from GPS positions recorded along the tracks. A principal component analysis was performed to provide a detailed description of the different trajectory patterns in relation with fishing trip specificities (i.e. vessel, gear, métier). Hidden Markov models were then fitted for some selected fishing trips. Two types of models were considered. The basic one was a 2 states model with behavioural activities corresponding to fishing and traveling. The second one presented a number of fishing states depending on the number of métiers identified for the trip. Fitting performance was compared based on DIC and estimated confidence intervals for the parameters. This procedure was applied to a set of volunteer vessels participating in the RECOPESCA project from IFREMER in the Bay of Biscay and the English Channel for years 2011-2012. We show that fishing trip activities, such as métiers, were structuring variables for trajectories, which helped to specify properly hidden Markov models. Discrete choice models building in a random utility function (RUMs) have also been used to aid understanding and modelling of fleet dynamics and to anticipate how fishing effort is re-allocated following any permanent or seasonal closure of fishing grounds, given the competition for space with other active maritime sectors. A first Random Utility Model (RUM) was developed and initially applied to determine how fishing effort is allocated spatially and temporally by the French demersal mixed fleet fishing in the Eastern English Channel. The spatial resolution of this investigation was that of an ICES rectangle (30’ x 60’). The explanatory variables chosen were past effort i.e. experience or habit, previous catch to represent previous success, % of area occupied by spatial regulation, and by other competing maritime sectors. Results showed that fishers tended to adhere to past annual fishing practices, except for the fleet targeting molluscs which exhibited within year behaviour influenced by seasonality. Furthermore, results indicated generally that maritime traffic may impact negatively on fishing decision. Finally, the model was validated by comparing predicted re-allocation of effort against observed effort, for which there was a close correlation. The method was also applied to the Dutch beam trawl fleet (2008-2010). The Dutch fleets’ activity was well captured by the model which included only biological and economic drivers. Predictions were accurate and followed the seasonal patterns well. To predict the long term changes in fishing activity additional factors, such as the competition for space with other marine users, should be included and changes in fish distribution should be linked to the current model. A second Random Utility Model (RUM) was developed using a finer spatial resolution (15’ x 15’) and initially applied to analyse the determinants of English and Welsh scallop-dredging fleet behaviour, including competing sectors operating in the eastern English Channel. Results show that aggregate activity, maritime traffic, expected costs, English inshore 6 and French sovereign 12 mile nautical limits negatively impact the choice of fishers, and conversely that past success, expected revenues and fishing within the 12 nautical mile limit have a positive effect on their utility. The model has potential application for Marine Spatial Planning (MSP). This RUM was also used to evaluate the interactions of fishing effort allocation and shipping for the Dutch demersal fleet fishing in the English Channel, this analysis of the French and UK fleets was also undertaken for Dutch seiners operating in the Eastern Channel. The parameters associated with the gross revenue all had positive parameter estimates for the means, as is expected from fishers seeking to maximise net revenues. The parameters associated with costs were also positive, which is striking, given that one would expect a cost minimization. The positive estimates could be caused by the trips to Dutch harbours that are in the data set. The closed area parameters are both negative, reflecting the fact that fishing is not allowed in these areas. The parameters associated with the shipping lanes had negative estimates, as in the French and English case, but these estimates did not differ significantly from zero. Other spatially-explicit statisti