General description of the tool

FAdvanced spatial analysis techniques and methods used to identify and manage the spatio- temporal nature of bycatch are acknowledged as being important in optimising catch composition and extending fishing opportunitis under the Landing Obligation (LO) (Dunn et al., 2011; Paradinas et al., 2016, Vignaux, 1996; Tidd et al., 2012; Van Putten et al., 2012; Vilela and Bellido, 2015). Survey data collected from research vessels, observer data collected from commercial fishing vessels and catch information from logbooks, coupled with VMS data can be used to produce maps that identify species hotspots. The inclusion of discards in such data sources provides more precise estimates of catch than just using landings data alone (Viana et al., 2013). Maps produced from observer data could, therefore, provide a real insight into the spatial distribution of all species caught by commercial vessels. There are, however, problems associated with discard sampling from observers including low sampling frequency and irregular sampling (Villasante et al., 2016).
The Celtic Sea contains stocks of several important ground fish species, thus supporting a number of international fleets including those from Ireland, France, and the UK. The collaborative nature of the DiscardLess project has meant for the first time observer data from vessels operating out of these three countries can be combined for the Celtic Sea. Such an opportunity helps to overcome some of the problems associated with the sparse nature of data supplied by observers. This work, therefore, aims to use a tri-national dataset to identify areas where catches of species subject to TAC (total allowable catch) are likely to occur within the Celtic Sea. The resultant maps can be used to identify hotspots of catches that fishermen may want to better target or avoid to optimise catches under the landing obligation. The resultant information will be presented in the form of an interactive app so that fishermen can extract tailored information, ultimately helping to inform where to fish to reduce bycatch.

Input data

Data

Data collected by onboard observers working on Irish, British and French vessels operating in the Celtic Sea between 2010 and 2015 were used in the analyses. Data were collected by each member state as part of the EU data collection framework (Council regulation (EC) No 199/2008). Observer data were specifically used as it is the only source of information on the component of catches that are discarded at sea providing biological data on the whole catch and not just fish and shellfish later landed. In addition meta-data such as the position and duration of hauls, gear and mesh size used, vessel type and vessel size are all collected by observers. Initial analysis concentrated on data collected on-board TR1 vessels i.e. those operating bottom trawls, Danish seine nets and similar towed gear with mesh sizes between 70mm and 100mm, but excluding beam trawls (Davie and Lordan, 2011).

Mapping Methodology

The geographical midpoint of all hauls were calculated and catch data assigned to this point. Catch data were then assigned to 0.2 by 0.2 degree grid cells to ensure individual vessel and national data could not be identified. The proportion of the haul by weight for both the below and above MCRS (minimum conservation reference size) component of the catch for each species subject to a TAC was calculated. Mean annual values were calculated for each grid cell and grid cells were subsequently binned into 20% intervals with an additional category being used to identify where grid cells contained zero catch within a year. A final, amalgamated map for 2010 through to 2015 was created for each species and size component grouping by identifying grid cells that were consistently and uniquely within the same binned category over multiple years (Fig. 1). The above process was also conducted using catch per unit effort (CPUE) rather than proportion by weight. CPUE was calculated by dividing the total weight of both below and above MCRS TAC species caught in each haul by the total haul duration. Again mean annual values were calculated for each grid cell and subsequent values were divided into five equal quantiles, following the removal of zero catches. Again an amalgamated map was created for the whole time period studied by identifying grid cells that were consistently and uniquely within the same quantiles over multiple years. In addition to determining annual catch patterns, seasonal patterns were investigated by sub setting the observer data into four data sets based on the quarter of the year in which fishing operations took place. Mean quarterly values per grid cell for each individual year were calculated before again being binned or assigned to quantiles with amalgamated quarterly maps being produced for both proportion and CPUE data as before.



Figure 1. Diagram showing the steps in the map production process. (A. Individual binned maps created for each year; B. Amalgamated map for all years identifying grid cells within consistent binned categories over multiple years; C. Final interpolated map)

Interpolation

The resultant gridded maps show where over time consistent proportions or volumes of certain species within the catch are likely to be found. This provides valuable information to inform where to target fishing activities and optimise catches in relation to available quotas. To provide a more user friendly end product the grided maps were interpolated using inverse distance weighting using gstat in R (Pebesma, 2004; R Core Team, 2012). Due to the grid structure of the data a number of suitable interpolation techniques were compared prior to the final interpolation technique being applied. Proximity polygons, nearest neighbour analysis and inverse distance weighting techniques were validated against a test data set and the root mean square error (RMSE) was calculated for each method (Luo et al., 2008). The inverse distance weighting interpolation consistently produced the lowest RMSE values for each interpolated map and thus this method was used throughout our analyses.

Discussion

Hotspot mapping provides essential information to allow the optimisation of fishing efforts to catch target species and avoid unwanted and quota restricted species. Observer data, collected from commercial fishing vessels, provides an ideal basis for such maps as these data include the discarded component of catches, in addition to landings. The sparse coverage of observer data and limited sampling of commercial vessels can present problems when trying to identify patterns in such data. By combining data from three EU member states with commercial vessels operating within the Celtic Sea for the first time, we were able to produce maps highlighting where catches of TAC species show consistent patterns over multiple years. Further by using the output of these maps in an interactive app we have produced a tool that can easily be used by stakeholders to help inform decisions on where to fish to reduce unwanted catches.

Two catch metrics were used to identify where similar catches are expected to occur over time. CPUE gives an indication of how the volume of a species in a catch varies. When trying to avoid non-target species it would make sense to avoid areas where there is increased probability of high CPUE catches. When targeting species, although stakeholders are likely to be drawn to areas with high CPUE, it is also important to consider the proportion of that species within the catch and how clean the catch is if bycatch is to be avoided. Thus it is important to use these two metrics together, depending on what is driving fishing behaviour and how restrictive other quotas may be in relation to the target. Our example comparing CPUE and proportion maps for above MCRS haddock shows how these two metrics compare and contrast. There is some agreement between the two maps as to where hotspots of this species occur, and identifying these areas would be beneficial to best target fishing. There is certainly less relief in the maps based on proportion by weight and this is especially true of the below MCRS component of the catch where often there is never greater than 20% of the catch by weight. For these cases the CPUE map provides more detailed information as data categories are based on quantiles rather than pre-defined equal intervals.

Under the LO all catch of TAC species regardless of size will count against quotas (European Commission, 2013). Thus it is important to avoid all below MCRS fish as this component of the catch cannot be sold for human consumption and receive full market value. Whilst the extent over which below MCRS catches are likely to occur may overlap with that of the above MCRS component of the same species we have shown that for whiting areas can be identified where it would be possible to target large, marketable fish whilst reducing the chances of catching smaller fish. In the Celtic Sea, where there is a mixed demersal fishery and numerous species co-occur it is also to be able to highlight those areas where fish are less likely to occur together to allow fishermen to target certain species whilst avoiding chokes. Haddock and whiting are two species that co-occur but for which there are often uneven quotas (Calderwood et al., 2016) and so it may be necessary to target one species whilst attempting to avoid the other. Again the comparison of maps for these two species showed that there is potential to concentrate fishing efforts in areas to minimise the likelihood of catching one whilst maximising the likelihood of catching the other. The mapping method adopted identifies areas with consistent proportions or volumes in the catch over time. Obviously fish populations are mobile and aren’t always going to be found in the same location. These maps do, however, give an indication of where the likelihood of catching certain species is greatest. Examining how cod distributions vary with season shows how important it is for stakeholders to consider the data provided in these maps at greater temporal resolution than just annually. Due to the resolution of input data used in this work, quarterly data is currently the finest temporal resolution that the maps can be divided into. This still provides greater detail than solely presenting annual data and allows fishermen to consider how the dynamics of fish stocks over the course of a year requires seasonal adaptation to fishing practices.

It is clear that these maps hotspot provide information that is essential if stakeholders are to make the most informed decisions when choosing where to fish whilst operating under the LO. A suite of measures from gear adaptations through to the provision of spatial-temporal information will be required for the fishing industry to successfully reduce unwanted catches and meet the legislative requirements of the LO. The data provided in these maps provides one element of this suite but to ensure this data is easily accessible and digestible it needs to be presented to industry in the appropriate format. As a result an interactive app was developed to allow stakeholders to pick and choose the information that is relevant to them at any one point in time. They are able to select those species they want to target as well as those they wish to avoid and display all of the relevant information on one map. Areas with a high chance of catching solely the target species can easily be identified and help stakeholders to make the most informed decisions when deciding when, where and how to fish to avoid unwanted catches. Making the information stored within these hotspot maps easily accessible could aid in making fishing operations in the Celtic Sea more efficient, ultimately reducing operating costs. By arming industry with such knowledge and information it is hoped fishing operations can be optimised with fisheries continuing to be profitable whilst operating under the LO.

Outputs

To produce a user friendly and interactive tool for use by stakeholders an app was developed using Shiny and Leaflet in R (Chang et al., 2017; Cheng et al., 2017). Layers were extracted from the interpolated maps based on the original bin and quantile categories. These were converted to spatially referenced shape files and saved separately. Users of the app are able to select the time period they are interested in (Annual Data, Quarter 1, Quarter 2, Quarter 3 or Quarter 4), the numerous species they wish to target and those they wish to avoid. For each species selected the user can specify whether they are interested in the below or above MCRS component of the catch. They are then able to toggle the levels of catch they either wish to target or avoid, selecting either the minimum proportion of the selected species or minimum level of CPUE of interest (Fig. 2). Multiple target and non-target species can be selected at once and semi-transparent map layers are displayed on an interactive map, identifying where selected levels of catch are likely to occur. Figure 2. A screenshot of the shiny app developed to allow stakeholders to select the size, species and quantity of fish they would like to target and/or avoid during different seasons. The resultant map displays layers representing where to target or avoid fishing operations to optimise catch composition.

Results

The maps as described previously have been created for the above and below MCRS component of the catch for all demersal species subject to a TAC. To better focus comparison and analysis of the results this paper will focus on three key species; haddock, whiting and cod. Both haddock and whiting have been recognised as high risk species in the Celtic Sea with catches exceeding TAC across multiple member states (Rihan et al., 2017). Cod has also been noted as being at moderate risk for member states as a whole but presents particular problems for the Irish fleet due to low quota share amongst its’ vessels (Calderwood et al., 2016). Examples of how the information in these maps compares and contrasts for selected species are described below.

CPUE vs Proportion

Both CPUE (kg hr -1 ) and proportion by weight in the catch were used as metrics to identify hotspots of key TAC species. Figure 3 shows an example of how these two metrics compare for above MCRS haddock in the Celtic Sea. Areas with consistently high levels of haddock CPUE within the catch are centred around the coordinates 51.1 -6.85 between the south coast of Ireland and the north coast of Cornwall and to the west of Ireland centred around the coordinates 52.5 -11.0 (Fig 3.A). Catches where above MCRS haddock consistently constitutes at least 20% of the catch are identified in similar locations (Fig3.B), although for both areas higher proportions of haddock are identified closer in towards the coast than with the CPUE data. There are also a few discrepancies with small hotspots of high CPUE areas at 51.4 -9.4, 51.4 -11.2 and 50.3 -10.5 not being reflected in the proportion data. Similarly areas with high proportions of haddock identified at 51.79 -10.49 and 51.13 -7.14 are not reflected in the CPUE data. Less relief is also evident on the map representing the proportion of haddock in the map as few catches were identified with greater than 60% of haddock in the catch. Figure 3. Interpolated maps identifying A. Areas with consistent levels of >MCRS haddock CPUE over multiple years (2010-2015) and B. Areas with consistent proportion of >MCRS haddock in the catch by weight over multiple years (2010-2015).

Below MCRS vs Above MCRS catches

All species maps were created for two size categories based on fish either below or above MCRS, allowing for a comparison of the distribution of these two size categories amongst and between species. When comparing whiting catches the largest volumes are again caught in an area centred around the coordinates 51.1 -6.85 (Fig. 4). The majority of points with the greatest CPUE of below MCRS whiting are also encompassed by the areas with greatest CPUE for above MCRS whiting.

Areas with high CPUE of below MCRS whiting do, however, cover a much smaller area compared to the above MCRS component of the catch. Small distinct hotspots, representing the highest CPUE category, cover a total area of less than 430km 2 for below MCRS catches compared to 9500km 2 for the above MCRS catches in the same CPUE category. There are also a few distinct patches identified as having a high likelihood of high above MCRS catches where there are no below MCRS catches identified. Namely along the 52.5 degree latitude line and just off of the south west coast of Ireland at approximately 51.3 -9.28. Figure 4. Interpolated maps identifying areas with consistent levels of whiting CPUE over multiple years (2010-2015) for A. Below MCRS fish and B. Above MCRS fish

Species Comparison

The same metric can be used to compare the likelihood of different species co-occuring. The area with consistent proportions of haddock in the catch over multiple years in the Celtic Sea for example is much greater than that of whiting (Fig.5). No areas are identified as having consistent proportions of whiting in the catch south of the 52.5 degree latitude line or west of the -9 degree longitude line (Fig.5B). There is a distinct chance of catching haddock in this area, with some hotspots of up to 60% of haddock being identified (Fig.5A). There is also overlap of the haddock and whiting map extents, especially within area VIIg. Overall there are relatively small areas being noted as consistently having at least 20% of whiting in the catch, whilst areas identified as likely to have at least 20% of above MCRS haddock in the catch cover a much greater extent (approximately 5200km 2 compared to 38700km 2 ). Figure 5. Interpolated maps identifying areas with consistent levels of the proportion of above MCRS A. Haddock and B. Whiting in the catch over multiple years (2010-2015)

Seasonal variation

Seasonal variation in catches can be identified by examining amalgamated quarterly rather than annual data. When comparing areas where there are likely to be consistent levels of CPUE of above MCRS cod over each quarter distinct seasonal patterns can be seen (Fig.6). In the first quarter of the year areas with consistent levels of cod in the catch are found in a small number of isolated spots within area VIIg. Moving on to quarter two there is a sudden large expansion in the range of the area where cod is likely to be caught. In this period the area covered by the map layers extends to most of area VIIg. During the third quarter of the year the extent of the map retracts a little, splitting into two smaller regions within VIIg and also extending further east into VIIf, with some hotspots being concentrated along the boundary between these two ICES areas. In the final quarter of the year the extent of the cod CPUE map retracts further shifting north towards the south coast of Ireland in addition to a small hotspot emerging just above the north coast of Cornwall in areas VIIf. Figure 6. Interpolated maps identifying areas with consistent levels of CPUE for above MCRS cod over multiple years (2010-2015) for each quarter of the year.

Shiny App

All of the maps produced provide useful information that can be compared in numerous ways depending on the user’s interests and objectives. Providing the maps in an interactive app thus provides the opportunity to pick those layers of interest to easily compare and contrast. Figure 2, for example, shows the overlap of four different layers. The first two are species that wish to be targeted. Both above MCRS haddock and whiting have been selected as the target species, with the level of CPUE being set to include the highest two levels identified during the mapping analysis. Below MCRS haddock and whiting are selected as the non-target species, again with the layers highlighting the two highest levels of CPUE identified. The resultant overlap of all of these layers is displayed within an interactive map layer. Although there are overlaps between all four of the layers there are distinct areas that highlight where just the target species are likely to be found.