Introduction

As human population growth continues to increase there is a need to balance competing demands for natural resources. Traditionally seen as a common property resource, the sea is confronted increasingly with competition for space by competing sectors, e.g. fisheries, oil and gas exploitation, aggregate extraction, wind energy, shipping and transport, recreation, dumping and the military activities. Spatial planning and the regulation of human activities and pressures at sea are therefore becoming a concern, especially given that some resources are limited in space and quantity. Since 2008, the European Union has placed a responsibility on member states to achieve common principles based on the “Roadmap for spatial planning” [1], which falls under the Integrated Maritime Policy (IMP; [2]), and is generally referred to as Maritime Spatial Planning (MSP). The objectives of MSP are to manage anthropogenic activities in space and time, precluding or minimising conflicts between competing sectors without negatively impacting the ecosystem, operating within the Marine Strategy Framework Directive (MFSD; [3]). However, because sectors at sea can change rapidly and the complexities of natural systems are linked and inter-reliant, a management decision for one sector may affect others, and MSP needs to be treated as a process of continuous, adaptive management process.

Given the importance of MSP, several writers have stressed the importance of fleet-based spatial management in the commercial fisheries sector [4], [5], accounting for different fleet activities at a scale fine enough to be integrated into the MFSD process. To date, integration has been difficult owing to the broad scale (ICES statistical rectangle ∼900 nautical miles²) at which some data (e.g. landings) are reported. With the emergence of Vessel Monitoring Systems (VMS) over the past decade, however, MSP is now potentially possible at a finer scale. Issues of data confidentiality between member states have hampered the use of this information, and there is also a historic reluctance of fishers to provide accurate landings information for fear of conceding knowledge of profitable fishing grounds [6], and that the information might be used against their interests by other authorities. For example [7] suggested that fishers are concerned that conservationists might identify productive fishing grounds as suitable for Marine Protected Areas (MPAs), or fisheries managers might implement tighter enforcement constraints. In the light of the limited data availability and confidentiality, fisheries managers are looking now for alternative approaches to assist spatial planning, which will reduce implementation error i.e. where the effects of management differ from that intended [8].

One such approach involves anticipating fisher behaviour in response to regulation. Fisher behaviour cannot be predicted with certainty because of the many factors which influence where and when a fisher will operate. However, if managers can better anticipate fisher behaviour, then they may be able to reduce the unanticipated side-effects of management actions aimed both at the fishery sector and at other sectors. Traditional fisheries management treats fishers as static and homogenous with no consideration of their behaviour and individual aims [9], [10]. Recent studies have applied random utility model (RUM) methodology [11]–[13] to this issue, because such models offer an opportunity to study individual behaviour at a finer scale of space and time than previous approaches [14].

The objective of the present study was to model the key determinants of where fishers choose to fish, building on retrospective time-series and including interactions with a selection of key sectors also competing for space in the area. In this study we acquired data for the English and Welsh scallop-dredging fleet operating in the eastern English Channel (ICES Division VIId) and these fleets form the basis of our case study. This area also contains one of the busiest shipping lanes in the world, the route between the Atlantic Ocean and the North Sea, which we hypothesise might have a negative impact on commercial fishers. There are also several active marine aggregate extraction sites and fishers have expressed concerns about the accumulation of such sites and the effect of fishing pressure concentrating elsewhere for fear of gear damage, and the sustainability of fish stocks that are already heavily exploited in these areas [15]. The fishing restrictions in the area consist of local English bylaws prohibiting beam trawlers of > = 300 horsepower or 70 gross registered tons from the 12 mile belt of sovereign waters around the English coast to restrict competition with the inshore sole fishing fleet (Fig. 1). This ruling also prevents fishing by any international fishing vessel, though the area can be used for safe passage. There is a further 6 mile restricted zone to assist inshore vessels by prohibiting some fishing vessels of size >14m (depending on which regional Inshore Fishery and Conservation Area—IFCA they fall under) and limitations on scallop vessels with a certain number of dredges. Most of the vessels operating in the region are small (<10 m) inshore boats that deploy gillnets, trawls, longlines, traps and pots, and target sole (Solea solea), plaice (Pleuronectes platessa), cod (Gadus morhua), bass (Dicentrarchus labrax) and some skates and rays (Rajidae; [16]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Spatial overlap of sectors (coloured pixels represent maritime traffic densities, thick and thin white lines the12 and 6 mile limits respectively, aggregate mining sites (labelled) and the dashed black line proposed Special Areas of Conservation) within the English Channel.

https://doi.org/10.1371/journal.pone.0116335.g001

A mixed logit RUM was developed to analyse the determinants of fisher behaviour at a fine scale using English and Welsh VMS data. This model evaluates the effect of the key potential competing sectors on fishing behaviour. Suggestions are then made as to how the method can be used in integrated MSP in anticipation of the potential establishment of Special Areas of Conservation (SACs) in the area as part of UK commitments to the EU’s Habitats Directive [17] or Marine Conservation Zones (MCZs) under the UK Marine and Coastal Access Act 2009.

Materials and Methods

The model

Having populated the dataset with all covariates, we developed a mixed RUM to determine the key determinants of fisher behaviour in relation to competing sectors and fishing-specific covariates. We hypothesise that key competing sectors of activity as well as fishing costs (i.e. fuel price) negatively impact the spatial coverage of fishing operations (as presented in Fig. 2), in contrast to expected vpue and past effort (knowledge or habit) which positively influence fishing operations. Pioneering research by [30], [31] on the use of discrete choice and economics methodologies demonstrated the relationship between utility maximization and discrete choice, where utility influences individual choice with a deterministic and stochastic error component. RUM derives its name from discrete utility maximization and assumes that the choices are random to the analyst. A mixed logit choice RUM was implemented because it relaxes the non-IIA (Independence of Irrelevant Alternatives) assumptions associated with preference heterogeneity among fishers. This approach is efficient in dealing with panel data for repeated individual choices, as is the case within this study. For a detailed explanation of mixed logit, see [32] and [33]. Succinctly, the total utility μ_njt of fisher n for site j in trip t is (1) where β′_n x_njt represents the observed utility and ϵ_njt is the error distribution that is part-correlated and part independently and identically distributed (iid) over alternatives and individuals[31], [34]. Within the mixed logit framework, β_n was assumed to follow a normal distribution, and for a given value of n (for simplicity disregarding t), the conditional probability of choice j across all other choices j = 1 to J is estimated by drawing random values β by simulation using (2) where β_n is a vector of coefficients that varies across individuals, and x_nj is a vector of the attributes of each of the choices made. All covariates met the normality assumption following log-transformation. The analysis on 3019 observations was carried out in the SAS package PROC MDC [35] using quasi-Newton optimisation and 100 Halton draws.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. The eastern English Channel displaying total annual scallop dredging effort densities in hours fished (black pixels = 200 hours).

https://doi.org/10.1371/journal.pone.0116335.g002

The definition of choice set

When designing RUMs, fisheries scientists are confronted with the problem of creating a choice set, which covers the individual sites to which a fisher travels to fish. If sites are too small (individual latitude/longitude positions), there may not be sufficient site-specific information, but if they are too large, important site-specific information can be lost when aggregating, losing information valuable to policy-makers. Fishers have prior knowledge of resource distribution and habitat [36], [37], and scallops are relatively static molluscs, suggesting that in future years, any choice set will be subject to relatively little change. On the basis of this assumption, the predetermined area making up the choice set for this study was based on the 2005–2010 effort distribution of scallop dredgers plotted from the VMS records (Fig. 2). A trade-off in scale was necessary so the dataset was grouped into 45 sub-rectangles, determining the choice set (Fig. 3). Having too many specific choices can reduce the stability of standard algorithms for maximum likelihood estimation as the number of alternatives rises past around 50 due to data multicolinearity [38].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. The eastern English Channel with ICES rectangles overlaid and the choice set represented by the hatching geo-referenced by ICES rectangle and the eight sub-rectangles within.

https://doi.org/10.1371/journal.pone.0116335.g003

Variable selection

Net benefit or profit per set of fishing trips is not easily computed as detailed cost data (variable and fixed) is costly to collect and such confidential information is not usually disclosed. Researchers therefore use a proxy of value per unit effort (vpue) rather than cost, which relates to the net benefit of variations in stock density [39], [11]. Value per choice was calculated as a proportion of the total value (revenues from landings) per ICES rectangle based on effort (hours fished) derived from the VMS, and vpue was then computable. The average vpue by year, month and location was calculated for the fleet and lagged both by month and annually to take account of spatial fluctuations at different temporal scales. The past percentage of a particular vessel’s scallop trips to a fishing location as a percentage of the fleet total elsewhere was used to represent fisher habit/experience and to track the seasonal nature of the fishery, as in [40], and was also lagged as above.

Where possible, fishers are assumed to maximise their returns [41]. Subject to the weather and other factors, they trade off travel costs against the quality of the fishing grounds. A proxy for perceived costs was calculated based on the average fleet distance to landing port from VMS fishing locations, calculated using the Haversine formula [42], weighted by mean average fuel price from fishing in the same location in the same month of the previous year fishing (i.e. lagged average costs). [43] survey of fishers in the south west of England showed that fishers routinely keep track of fuel prices in order to forecast their potential earnings after deductions for other costs. Landing port was used as it was assumed that the fishers would have prior knowledge of seasonal market prices in the proximity of fishing locations.

Aggregate mining activity enters the model as the average percentage coverage of this mining activity in the location the previous month (to capture potential past activity as a nuisance to fishing operations). The 6 mile limit (as a proxy for the English restricted zone for certain vessels over 14m) and the 12-mile limit (as a proxy for the French internationally restricted zone) were treated as a spatial constraint. Maritime traffic was included as average hours in which a location was occupied by marine shipping traffic in the previous month. Finally, as a proxy for congestion and social influencing effects, we included the average hours fished the previous month by English, French and other (unidentified fishers grouped) fishing vessels. The variable selection set was merged with individual scallop trip data by year, month and location, such that for every trip, the decision-maker had a choice of the specified 45-subrectangles (Table 1). Based on the historic time-series of VMS data, fishing activity was ascribed to a particular sub-rectangle, and values took a value of 1 if a location was selected or 0 otherwise It is important to note that for any particular vessel and any given trip, a number of observations may exist in a number of different sub-rectangles, hence each choice is considered separately as components within a fishing trip).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Definition of variables used in the RUM to model fisher location choice for the 45 ICES sub-rectangles in the eastern English Channel as defined in Fig. 3.

https://doi.org/10.1371/journal.pone.0116335.t001

Results

The mixed model showed a McFadden’s pseudo-R ² of 0.19, suggesting a very good fit [44]. Theoretically, the range for McFadden’s pseudo-R ² is between 0 and 1, but the general rule of thumb is that any value from 0.2 to 0.4 suggests an excellent fit, comparable to an ordinary least squares (OLS) R ² of 0.7–0.9 [45].

All mixed model coefficients were statistically significant (p < 0.01) except the coefficient for the average vpue of scallop from fishing in the same location in the same month as in the previous year and the proxy for congestion/social influence in the previous month of the current year for the French fleet (Table 2). The estimated standard deviations of the estimates were not significantly different from the mean (indicating that the parameters do not vary significantly in the population of fishers) for past vpue, cost, average percentage coverage by marine traffic and average hours occupied by fishing activity by English/other fishing vessels. Conversely, the effort to the location in the previous month in the current year, the average percentage coverage by aggregate activity and the average effort in the same month the previous year did vary, perhaps relating to variations in characteristics of the fishers not captured in the model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Estimated parameter values, where the dependent variable took a value of 1 if a choice was made or 0 otherwise.

https://doi.org/10.1371/journal.pone.0116335.t002

The effort distribution maps in Fig. 2, coupled with the model results (Table 2) show how the scallop dredges interact with the shipping traffic separation scheme, aggregates and fisheries outside the English 6 and French 12-mile limits. In general the mean coefficients show the signs one would expect: English scallop fishers are negatively affected by the English 6 and French 12 mile restrictions, as well as aggregate activity and marine traffic. Despite this, in every year of the study there was a large amount of fishing effort in these areas, even more so in 2010 within the high-traffic area, perhaps because of a trade-off with larger expected vpue in these areas. There is a significant positive influence of vpue in the previous month on the tactics of fishers, but not of vpue in the same month of the previous year, which strongly suggests in-year variability is the key driver of behaviour. In contrast, cost was a negative influence as expected. Past effort variables, which were included to depict habit or knowledge of past success of fishing grounds, have positive coefficients, suggesting they are important drivers of fisher location choice.

As part of a ‘what if analysis’ a series of numerical simulations revealed that fishers responded to a 50% decrease in % area covered by aggregate extraction differently depending on the spatial position of the activity. In an area close to shore associated with aggregate extraction (30E9G), there was a relatively large difference in probability of fishing, of +0.012, while there were small noticeable increases in nearby areas 30E9F and 29F0C. Doubling the coverage of aggregate extraction resulted in fishers moving out of these locations (notably these same areas, 30E9G, 30E9F and 29F0C) with a change of probability of −0.018, −0.012 and −0.004 respectively. Fishers moved into a more offshore site of existing extraction (29F0B), which recently has contained very high fishing effort, with a change in probability of +0.014. These observations suggest that aggregate mining activity heavily influences fisher decision making, possibly due to knowledge of the habitat that scallops live in, coupled with past experience (Fig. 2).

Most of the main scallop grounds are located within busy marine traffic areas (Fig. 2) and therefore one would expect that with a decrease in traffic intensity there would be less competition for space and fishers would move into these areas. Maritime traffic, however, showed relatively small effects. Doubling the coefficient of maritime traffic intensity resulted in fishing effort being displaced out of the traffic lanes, essentially spreading out, whereas halving the coefficient led to an increase in predicted effort into the traffic lanes, most notably 29F0A, 29F0B and 29F0C. Changes in expected fuel cost did not result in large significant differences in probabilities of site choice. When fuel prices are halved fishers move closer in shore to the English ports, in contrast when they are doubled fishers move to areas offshore where the concentration of fishers and expected vpue is at its highest (e.g. areas, 29F0B, 29F0C and 29F0D) or nearer to French landing ports, resulting in a ‘complimentary effect’ with expected costs and expected vpue (Fig. 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Changes in probabilities when halving or doubling each variable in contrast to the benchmark model’s observed variable mean values.

https://doi.org/10.1371/journal.pone.0116335.g004

Discussion

It is widely recognised that decision-makers and managers desire an ecosystem-based approach to address interlinked drivers of social well-being [46]. Marine Spatial Planning necessitating the balancing of multiple objectives; fisheries managers need to understand the implications of effort displacement from closing an area and the unforeseen consequences of their management actions (e.g. effects on other marine life, economic implications and effects on other maritime sectors). Several authors have stressed the importance of anticipating fisher behaviour in response to management regulation, in order to reduce implementation error [47]–[49]. Here, a mixed logit RUM was applied at fine-scale resolution to assess the key determinants of scallop fisher behaviour in the eastern English Channel, so that if a regulation or new activity emerges, fishing effort re-allocation can potentially be predicted.

A key finding was that past fishing success in a location within the previous month was a good predictor of continued fishing in that location. This can be interpreted as a proxy for habit, knowledge or experience, as in other studies [40], [10]. Similarly, the expected marginal net revenue of visiting one fishing site rather than another, in terms of vpue, was significant as expected [50]. This is more apparent for the vpue in the previous month, rather than in the same month the year before, potentially capturing either seasonality or more likely short-term temporal correlations in stock abundance (see Table 2). Surprisingly, perceived fuel costs were not a major driver in choice of fishing grounds, possibly because of the proximity of grounds to landing ports in the eastern English Channel. The French12-mile limit and English 6 mile limit unsurprisingly had negative influences on fisher site choice, possibly because of productive fishing grounds within limits, which are rendered unavailable to the scallop vessels. Nevertheless competition from the inshore national fleet could become an issue if the fleet is forced to occupy a reduced spatial geographic footprint than was previously the case, for example by spatial closures. Of further policy importance are the effects of other commercial maritime activities (e.g. transport, aggregates mining) on the behaviour of the scallop fleet. If interactions with these sectors are better characterised then the implications for the scallop fleet of other maritime sectors can be assessed in advance.

The analysis indicates that this fleet exhibits some risky behaviour in their responses, as the mean of the coefficients determining site choice and the estimated standard deviation of the coefficients in Table 2 show highly significant estimates of some of the drivers, suggesting that the parameters vary within the wider population of fishers [40]. The signs of the standard deviations in some instances are negative, but for estimation purposes they are free to take any sign, because the normal distribution is symmetrical around its mean, and the absolute value can be taken to estimate the variance. For the coefficients that do vary between fishers (i.e. previous effort (last month and last year) and presence of aggregate extraction) we can assess what proportion of the population of fishers see these factors as positive or negative when making decisions about site choice. Taking the mean and standard deviation together the point estimates of the coefficients can be calculated by standardising the scores (z scores) and thus a probability can be calculated. The model suggests that 88% of the population of fishers see effort in the past month as a positive inducement to fishing in the same location again and 12% see it as a negative inducement, which will be dependent on the time they spent in the location previously and their success in terms removing the harvestable biomass, if they ‘fished out’ an area they are unlikely to return. Similarly past effort in the same month of the previous year is a positive influence on location choice for 58% of fishers. The areas occupied by aggregates mining are chosen more than expected with about 40% preferring fishing in these areas, in contrast to the other 60% seeing it as a negative influence, confirming the assumption that the aggregate industry does impact scallop fishing. [15] found that by setting aside marine areas for aggregates mining, this resulted in reduced fishing effort. Since 2005, aggregate extraction licences have been granted over large areas [51]. This is contrary to [52] findings for sole, which suggests that aggregate mining can have a positive effect on the catchability of sole by beam trawlers and hence on profitability. Perhaps, increased turbidity increases sole catchability (by reducing visual cues for escape and/or fish being disturbed from the seabed) or the dispersal of food into the water column encourages sole to move away from the bottom to feed or they may favour the previously mined area because of changed food resources or substrate. In stark contrast however, a recent study by [53] using time series cross correlation approach concluded that aggregate extraction activity, proximity and intensity didn’t have any impact on fisher activity. The differences could be attributed to the different statistical approaches employed. In the approach adopted by [53], only one sector of activity was investigated, in contrast to this study whereby two competing sectors were studied with the inclusion of economic data. ARIMA models (such as those employed by [53]) do not take account of individual behavioural interactions and are purely based on past time series behaviour, however they remain an excellent tool to support expert judgment.

The shipping Traffic Separation Scheme (TSS) in the English Channel controls one of the busiest shipping lanes in the world and attempts to mitigate against the possibility of maritime accidents, but can also impede fishing. The output from the model suggests that the presence of a TSS significantly reduces the probability of a fisher choosing a location, suggesting that the policy is having the desired effect of separating fishing from other activities, though at the cost of reduced ability to choose areas of potential high profitability.

This study gives clues to policy makers about the likely impact of their actions on fisher behaviour. For example, an increase in traffic densities would have a high chance of displacing effort to local inshore waters (Fig. 4). Conversely, the fleet responds to higher fuel cost by going further offshore with the expectation of the reward of higher returns, and when costs are lower they fish equally between inshore and offshore locations as they are not forced to cover higher costs by fishing in areas with highest vpues. Fig. 4 suggests that when aggregate mining is doubled there is a greater increase in fishing activity offshore and when halved there appears to be movement of fishing into the location of extraction. Also of note is the movement of vessels towards the French 12 mile limit, resulting in shorter distances to land to French ports.

A further important observation is that if one of the parameters that disadvantages fishers’ (e.g. shipping traffic densities) is altered, the competition for space effectively increases and the fishery spreads out. This may be because the traffic lanes are home to the best scallop fishing grounds and the specific location a vessel relocates to is the one with the next best trade-off between expected catch rates and distance to landing ports. This is also apparent for the competition with the aggregate sites, which are located in the heart of good scallop fishing grounds. Any reduction of the space taken up by aggregate mining, especially inshore results in an increase in effort allocation to those locations. This “fishing for space” where observed, could be viewed as symptomatic of competition within the fleet as well as a response to other sectors, and hence this could be used as a direct measure of spatial conflicts.

Conclusions and Future Work

The Eastern English Channel is a shared resource and there is increasing competition for space and resources, requiring novel management approaches that account for all or some of the interactions between sectors. To our knowledge, no other study has used a mixed RUM at fine resolution to assess key determinants of human behaviour in relation to different maritime sectors and as a possible tool for MSP. The results are promising and lay the foundations for future work that could include adding Marine Conservation Zones (MCZs) to the model. Final decisions on where MCZs will be introduced in the English Channel and what activities will be excluded are still to be clarified or have only recently been resolved, so it was not appropriate to incorporate simulated closures into the model. Nevertheless, the approach taken could be applied to other fleets, as RUMs offer the capacity to model individual behaviour at fine spatial and temporal scales needed for assessing the implications of policy decisions [54]. It would also be desirable to re-fit the data to recent data where fishing effort is more stable (during the investigated time period effort gradually increased) and as such results could appear somewhat different. Further work might include the Scottish fleet which represent a large proportion of scallop fishing effort in the eastern English Channel, evaluating trade-offs with both socio-economic and conservation objectives using efficient and effective spatial planning tools such as Marxan and MinPatch.

Acknowledgments

The research leading to these results has received funding from the European Union’s Seventh Framework Programme for research, technological development and demonstration (FP7/2007–2013) within the Ocean of Tomorrow call under Grant Agreement No.266445 for the project Vectors of Change in Oceans and Seas Marine Life, Impact on Economic Sectors “(VECTORS)” and time under Cefas contract (Seedcorn 67100G). We thank Andy Payne and Jim Ellis (Cefas) for their constructive advice and inputs to the work and paper, the UK’s Department of Energy and Climate Change (DECC) for providing data on fuel price, Royal Haskoning for supplying the commercial aggregate data and the Marine Coastguard Agency (MCA) for the AIS maritime traffic data. The paper is a contribution to Imperial College’s Grand Challenges in Ecosystems and the Environment initiative.