Mapping fine-scale demersal trawl effort for application in ecosystem assessment and spatial planning

Fine-scale maps of fishing activity are valuable information layers for fisheries management, assessments of biodiversity impacts and marine spatial planning. Our aim was to develop an accurate map of demersal trawling intensity in South Africa and to demonstrate its utility at a national scale. We calculated a swept area ratio, representing demersal trawling effort for the entire study period (2005‒2018) and annually. We then plotted spatial and temporal patterns of trawling activity, identified core fishing areas, and examined spatial overlap between trawling, South Africa’s marine ecosystem types and the national network of marine protected areas. A high proportion of trawling effort (90%) was concentrated in 43% of the area exposed to trawling, with the remaining 10% spread across 57% of the fished areas. The fishery overlaps with 33 of 150 benthic and bentho-pelagic marine ecosystem types. Of those, 11 have more than 50% of their extent, and five have more than 80%, within the trawl ring-fence. Our analyses support a systematic prioritisation of ecosystem types for further management and protection. The new South African trawling-intensity map contributes an improved pressure layer for ecosystem assessments, can help identify priority fishing areas and has application in conservation, marine spatial planning and fisheries management.


Introduction
Mapping fine-scale demersal trawl effort for application in ecosystem assessment and spatial planning JC Currie 1,2 * , LR Harris 2 , LJ Atkinson 3,4 , TP Fairweather 5 and KJ Sink 1,2 Amoroso et al. (2018) mapped trawling intensity in 24 regions across the globe from high-resolution vessel monitoring system (VMS) and logbook data.In South Africa, they used commercial vessel logbook start and end coordinates for the period 2008-2013 to calculate a swept area ratio (SAR) on a 1-km 2 grid, which estimates the proportion of each grid cell area covered by trawl drags annually.To achieve it from logbook records, they apportioned the trawl distance (derived from trawl duration and speed) to grid cells in a straight line between the start and end positions and used trawl-door widths estimated for different classes of vessels (Amoroso et al. 2018).This method provides an intuitive measure of trawling intensity and is easily translated into an estimate of trawling frequency, in that the period between successive sweeps of the grid cell area is represented by 1/SAR years (if calculated on a 1-km 2 grid).
There have been multiple efforts at mapping trawl activity in South Africa.The first attempt to compile a national trawl map was most likely that of Scott (1949), who mapped the grounds used by commercial trawlers during the first half of the 20th century.Although the maps do not capture the frequency of trawling, they contribute a valuable record of where trawling occurred during its first few decades of expansion in South Africa.Most of these areas remain key fishing grounds (Sink et al. 2019).
Prompted by conditions of the successful 2004 Marine Stewardship Council (MSC) eco-certification of the hake trawl fishery, the South African Deep-Sea Trawling Industry Association (SADSTIA) commissioned a study of the distribution of trawling effort and its interaction with sediment types (Wilkinson and Japp 2005).To map trawling intensity, Wilkinson and Japp (2005) overlaid three sources of information, namely: (i) a measure of trawling effort (hours; 2002-2004) recorded to 20-min grid cells in commercial logbook records; (ii) start and end positions of trawl events from an observer programme that covered ~10% of commercial vessel trips (July 2002to July 2005); and (iii) accumulation of trawl tracks recorded by marine navigation software (MaxSea) on commercial vessels.Using these information layers and assumptions about gear (trawl door) widths, they calculated a trawling-intensity index that estimated the proportion of trawled area swept annually at a 20-min grid resolution.
Following this work, and in response to considerations of fishing impacts on benthic habitats as part of ongoing MSC certification, SADSTIA undertook to voluntarily ring-fence their demersal trawl footprint and restrict it to areas previously fished by the industry, up to and including 2007.Wilkinson and Japp (2008) mapped the ring-fence on a 1-min grid, using navigational trawl tracks from the industry fleets and observer trawl-position records.The industry committed to not expand beyond this trawling ring-fence without a comprehensive environmental impact assessment, and the area limit is prescribed in the demersal trawling permit conditions issued annually by the Department of Forestry, Fisheries and the Environment (DFFE).Although the ring-fence does not represent trawling intensity, it plays an important role in constraining the extent of demersal trawling and is therefore important to investigations of trawling distributions.
A measure of demersal trawling pressure was mapped in the South African National Biodiversity Assessments (NBAs) of 2011 and 2018 (Sink et al. 2012(Sink et al. , 2019)).In NBA 2011, the number of trawl events recorded in commercial logbooks for the period 2000-2008 were summed within 20-min grid cells (representing ~1 141 km 2 at 34° S) and normalised to represent a relative trawling-intensity value of between 0 and 1.That coarse resolution was greatly improved in NBA 2018 (Sink et al. 2019), enabled by an initiative to record GPS start and end locations in commercial vessel logbooks, starting in 2004.In this case, the start and end coordinates of trawl drags were used to map the summed point density of trawl events within a 2.5-km radius of 120-m grid cells.The point density estimates were then re-scaled to a relative trawling intensity of between 0 and 100, with the lowest 10% cells removed to filter erroneous location records spread across areas not routinely trawled (Sink et al. 2019).
Although the trawling-intensity map generated by Sink et al. (2019) was a great improvement in the spatial resolution of trawl activity, the smoothed point-density approach is difficult to interpret other than as an index of relative trawling intensity.The SAR metric calculated by Amoroso et al. (2018) overcomes this limitation.However, closer examination of the resultant maps from both those studies revealed remaining artefacts from underlying errors in the logbook data.These included rounding errors of location data causing gridded patterns of concentrated effort and patches of apparent trawling activity outside of known trawling grounds.Therefore, we undertook to improve the mapping of demersal trawl activity in South Africa by applying the same approach as Amoroso et al. ( 2018), but to an expanded time-series of data (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) and with renewed emphasis on filtering or correcting erroneous position records.Our aim was to develop an accurate and intuitive, high-resolution map of trawling intensity and then to demonstrate the utility of such a layer in spatial assessments and planning.Objectives included: to rigorously clean the underlying records and develop a high-resolution and interpretable map of demersal trawling intensity; to investigate recent temporal and spatial patterns in demersal trawling activity; and to illustrate the potential application of the new trawl information layers in spatial assessments and decision support by examining interactions between trawling intensity, protection level, and South Africa's 150 benthic and bentho-pelagic marine ecosystem types (Sink et al. 2023a).

Data acquisition and preparation
Commercial trawl logbook records for the entire fleet (including inshore, deep-sea and midwater sectors) were requested from the then Department of Agriculture, Forestry and Fisheries (DAFF, now DFFE) for the period 2004-2018.Starting in the early 2000s, GPS coordinates were required as part of logbook returns, whereas the previous spatial reference of fishing records had been 20-min commercial grid cells.A substantial proportion of 2004 data lacked the GPS location records, so that year was excluded from analyses.Records were further restricted to demersal trawls by the inshore and deep-sea sectors (excluding pelagic or midwater trawling efforts).The remaining data represented the entire demersal trawling (and towed bottom-fishing gear) effort in South Africa during 2005-2018, consisting of 673 037 trawl events.All data preparation and analyses in this study were performed using R 3.6.2(R Core Team 2020) and RStudio 1.2.5 (RStudio Team 2020).Substantial errors were observed in the trawl location data, including coordinates that were in unrealistic locations (e.g. on land or beyond existing trawling depths) or concentrated in artificial, gridded patterns at multiple spatial scales (Figure 1).These included a tendency for points to be concentrated on unit degree lines (e.g. the prominent vertical line of points on 18° E) and on the vertices of 20-min grid cells (to which commercial trawls were previously recorded).Closer examination revealed a lower density of points in the 10-integer-min locations below the integer degree (e.g. the 10-integer-min locations to the left of the prominent 18° E line of points).A third pattern, obvious at higher resolutions, was that the majority of position records are gridded on (recorded to) integer minute locations (Figure 1), indicative that their coordinates were rounded to the closest minute.The last pattern or bias revealed during inspection of trawl locations was that there were notably few trawls recorded between the 59th minute and the next integer degree (e.g. between 17°59′ and 18° E) and a concentration of points on the 59th minute (i.e.location points had been rounded down to the 59th minute).These same patterns were noticeable both in latitudes and longitudes, and both in start and end locations (not shown), indicating systemic causes.
The errors in the distribution of trawl locations were carefully examined and corrected, or else filtered from the dataset if they could not be corrected.The 10-step process to do so is briefly outlined below, while details of the changes made and illustrations of the corrected errors are provided in Supplementary Materials S1.Corrections were based on the distribution of unaffected records and applied in a sequence of steps that sought to avoid them being informed by additional underlying errors not yet corrected.Hence, 1) trawl records that lacked coordinate positions for either start or end locations were removed; 2) a small proportion of obvious errors were corrected manually; 3) trawl records of unrealistic length (distance) between start and end positions were removed, where unrealistic was defined as zero (identical start and end positions) or if the trawl distance was >55.6 km (30 nautical miles, in accordance with Amoroso et al. 2018); 4) trawl events that crossed from one corner to the opposite corner of 20-min commercial grid cells were removed, as those locations had been retrospectively added to records lacking GPS positions; 5) any records that had a start or end location outside of South Africa's mainland maritime domain (shore to outer edge of the exclusive economic zone [EEZ]) were removed; 6) the concentration of location records rounded to the 59th minute were randomly redistributed between the 59th and 60th minute; 7) integer-degree records that had been rounded from the preceding 10-min locations were redistributed to those locations based on the pattern of records not affected by the rounding error; 8) rounded integer minute values were distributed into the minute space that followed, by adding random deviates between 0 and 59 seconds; 9) trawl records that had a trawling distance far exceeding their expected maximum distance, based on trawl duration and a maximum speed threshold, were assumed to contain errors and were removed; 10) lastly, a density filter was applied to remove remaining error locations dispersed across broad areas, including many areas known not to be trawled.The density threshold filtered records of ≤4 (start or end) locations per km 2 -grid cell over the 14-year study period.
Although similar errors may be encountered in other studies, the combination of systemic biases encountered in this dataset is expected to be unique, as it relates to multiple events and procedures that took place in the historical recording, digitisation and database workflows.The most recent data (towards the end of the study period and thereafter) do not contain the described rounding biases, which will preclude the need for most of these corrective measures when mapping trawling intensity from more-modern data.

Calculation of swept area ratio (SAR)
The measure of SAR is the area swept by trawl gear over a defined period (usually a year) divided by the seabed area at a defined spatial scale (Amoroso et al. 2018).SAR was calculated on a 1-km 2 grid using the cleaned dataset, as detailed by Amoroso et al. (2018).The area swept within a grid cell was calculated as the product of trawling time, towing speed and dimensions of gear components sweeping the seabed (Eigaard et al. 2016).Trawl door widths (ranging from 60 m for inshore vessels to 110-185 m for offshore vessels) were obtained from a company and vessel-type specific reference table compiled by DFFE.To apportion the effort of each trawl drag to underlying 1-km 2 grid cells, a straight line connecting the start and end positions was divided into 600 segments and the effort distributed among the cells according to the proportion of segments that fell within the cells (Amoroso et al. 2018).Using a larger number of segments was limited by computational memory.Although the length of trawl drags varied, the segment lengths tended to be approximately 23 m (typically ranging from 4 to 58 m).SAR was estimated for annual periods and as average values for the entire 2005-2018 period, representing mean demersal trawling activity over 14 years.
To avoid underestimating the intensity of trawl activity because of the substantial number of erroneous records removed during data preparation, the calculated SAR values were upscaled to account for the previously removed effort (Table 1).Proportions of effort were measured from the duration of trawls (hours).There were two categories of fishing effort removed: (i) records that had at least one of their coordinates within coarse 2° grid cells that overlapped with the SAR area, with the effort from these records spatially reinserted by calculating the proportion removed from each 2° grid cell and upscaling the SAR values in that grid cell by the removed proportion; and (ii) trawls that were missing coordinates, or else both their start and end coordinates appeared to be spurious (outside of the SAR area), thus precluding spatial allocation of their effort.The effort from these records was reinserted as a 'global constant' by upscaling the entire SAR layer by the relevant proportion of removed effort.In this way, fishing effort that had been removed during data cleaning was effectively added back into similar areas of the SAR map, or used to upscale the SAR layer as a whole if it could not be spatially allocated.The SAR calculation and the upscaling step that followed were applied separately to annual periods and to the entire (2005-2018) dataset, resulting in the final 'upscaled' annual estimates of SAR and the 2005-2018 mean.

Statistical analyses
Statistics of SAR were calculated across years.A measure of how variable interannual catches were, relative to their mean, was provided by the coefficient of variation (CV), as CV = SD / x , where SD is the standard deviation and x − is the mean.Areas expected to be of importance to the fishery, in that they were fished intensively (high SAR values) and consistently (low CV), were highlighted with a priority index (P) created by multiplying 1 CV with the mean SAR ( )  x , which is equivalent to: (1) The summed fishing activity (SAR) was plotted over time and against the area occupied (number of grid cells that contained positive SAR values).To highlight the spatial areas that may have experienced increases or decreases in trawling intensity during the study period, slopes of linear fit were plotted on a map if they were different to zero (p ≤ 0.05), by fitting models to annual SAR values at each grid cell, using the package PolyTrend (Jamali and Tomov 2016).

Trawling-activity interaction with marine ecosystem types
To demonstrate the application of the trawling-intensity map in the context of spatial prioritisation and planning, we examined how trawling activity interacted with South Africa's marine ecosystem types (Sink et al. 2019(Sink et al. , 2023a)).We did so by quantifying the spatial overlap between affected ecosystem types, the trawl ring-fence (Wilkinson and Japp 2008) and the trawling-intensity map.We then examined the current representation of affected ecosystem types in the national network of MPAs, which was considerably expanded in 2019 (Sink et al. 2023b).
We report the areas and proportions of ecosystem types that are trawled, as well as the mean SAR intensities they are exposed to.These results map the distribution of trawling intensity over the ecosystem area and illustrate the cumulative area impacted by different intensities (SAR), as well as the ecosystem area (proportions) within and outside of the trawling ring-fence and MPAs.Although maps were plotted using geographic coordinates for ease of reference, all area calculations were performed using the Albers equal-area conic projection with standard parallels optimised for the study region (details provided in Supplementary Materials S2).

Spatial and temporal patterns in demersal trawling intensity
The trawl-intensity (SAR) map from this study provides the most accurate spatial representation of recent demersal trawling activity based on available data (Figure 2a), without the many artefacts from erroneous locations in the original trawl records (Table 1; Figure 1).The only discernible artefacts that remain consist of some straight lines (connecting start and end positions) in certain localised areas where vessels are known to trawl in a curve to avoid a feature or when following depth contours, or where they bridge across trawl grounds (evident offshore of Algoa Bay in Figure 2a), with either the start or end trawl location containing an error that placed it within another nearby trawl ground.Although visible on the map, these remaining artefacts represent only a small proportion of the trawling intensity.The maps of mean SAR for the study period (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) and for the annual periods are freely available online (Currie et al. 2021).
The map of mean trawling intensity indicated that 55 249 1-km 2 grid cells were exposed to trawling activity during the 14-year study period, which equals 5.2% of South Africa's mainland maritime domain and 78.8% of the trawl ring-fence.Annual means of the equivalent metrics were lower, namely 40 283 1-km 2 grid cells and 57.4% of the ring-fence.The map shows a concentration of fishing activity predominantly near the western margin of the continental shelf (including the upper slope) and in similar depths offshore of Algoa Bay (Figure 2a), in an area referred to as the Chalk Line grounds.Substantial areas of the demarcated trawl ring-fence appear to have remained untrawled during the study period (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018).A few small areas exist where the SAR layer extends outside of the ring-fenced area.The majority of these are related to the artefacts from the assumption of straightline trawls that cross concave edges of the trawl ring-fence boundary.Investigating relative variability among years, using the coefficient of variation (CV), shows that most pronounced variation tends to be in areas that are relatively lightly trawled, usually on the periphery of the fished areas (Figure 2b).Heavily utilised areas tend to have low CV values as they are consistently fished among years.
Investigation of summed annual SAR values, as a proxy for trawling effort over time, showed a steep decline in trawling activity between the highest level in 2005, and the lowest levels in 2009 and 2010 (approximately half of the 2005 level) (Figure 3a).After 2010, there was a partial recovery until 2016, after which fishing effort declined again in the last two years, ending at a level ~42% lower than the 2005 level.Plotting the summed SAR proxy of effort against the size of area used by the fishery shows a clear positive relationship (Figure 3b), indicating that years of greater fishing activity resulted in expansion of the geographic areas trawled.
Inspection of the spatial distribution of temporal trends in the 2005-2018 SAR time-series revealed where trawling activity increased or decreased during the study period (Figure 4).As might be expected from the decreases in effort depicted in Figure 3a, the slopes of linear fit to the annual SAR values suggest trawling intensity has declined in many of the fished areas, both on the west coast and on the Agulhas Bank.The most-prominent decline in intensity  was on the Chalk Line grounds south of Algoa Bay.There are correspondingly far fewer (and smaller) areas where trawling intensity has increased, although this is evident in localised areas, most notably the shelf edge area south and south-south-east of False Bay (Figure 4).Using 1/CV as a metric of how consistently areas are trawled among years, and multiplying it by the mean trawling activity (SAR), provides a measure to identify areas that tend to be consistently fished at higher levels and may be useful to identify core fishing areas across a multi-year period (Figure 5).Areas that score higher using such a priority metric tend to be well captured by an outline that incorporates 90% of the mean SAR values from the study period (Figure 5), which covers only 28.6% of the trawl ring-fence.
Trawling-intensity values (SAR) were highly skewed in their distribution (Figure 6a), with a large proportion of values at the lowest end of the range and with relatively few records with higher SAR values, accounting for the concentration of effort seen in Figure 5.The high proportion of low SAR values results in a relatively low average (3.15) over the 14 years and across all depths.The highest trawling intensities tended to occur between depths of 170 m and 700 m, and the highest mean values between depths of ~240 m and 630 m (Figure 6b).A few grid cells were exposed to intensive trawling activity in certain years.For example, a SAR value of 36.5 is equivalent to the area of a grid cell being swept by trawl nets every 10 days, which was exceeded 433 times in the annual estimates of SAR between 2005 and 2018.The highest mean SAR value over the entire study period was slightly less at 33.9, but 251 1-km 2 grid cells were completely trawled every 20 (or fewer) days, on average, during the study period.
The highly skewed distribution of annual SAR values (Figure 6) means that relatively large areas exposed to trawling were fished at relatively low levels; for instance, 59.4% of areas fished between 2005 and 2018 had annual SAR values of <2 (Figure 7a) and 45.6% of fished areas were exposed to trawling once a year or less, on average.A different way of looking at this is that a high proportion of effort (90%) was concentrated in a relatively small proportion of area (43% of fished area and 29% of the ring-fence), while the remaining 10% of effort (lowest SAR values) was spread across 57% of the area fished (Figures 5, 7b).

Trawling interactions with marine ecosystem types
Of the 33 marine ecosystem types affected by demersal trawling in South Africa, 21 had more than 20% of their extent within the trawl ring-fence (Table 2); 11 ecosystem types had more than half their extent within the trawl ring-fence, of which one was entirely within the ring-fence, while four more had >80%, and three others had between 70% and 80% of their extent within the ring-fence.
The calculation of ecosystem-type areas that were covered by (non-zero) SAR values provides an estimate of areas trawled during the study period, as opposed to the 'ring-fence', which represents areas where trawling is permitted but where some proportion of the areas may remain untrawled for certain periods, depending on the behaviour of the fleet.For most ecosystem types, the values for these two area metrics were relatively similar, indicating that the trawlers were utilising most of the trawl ring-fence area.Exceptions, however, included that of the Southern Benguela Sandy Outer Shelf, where ~4 617 km 2 of the ring-fence appeared to remain untrawled during 2005-2018.Similarly, 3 786 km 2 of the Southern Benguela Outer Shelf Mosaic, 2 087 km 2 of the Southeast Atlantic Upper Slope, and 1 090 km 2 of the Southwest Indian Upper Slope were within the trawl ring-fence yet appeared to remain untrawled during the study period.
A measure of trawling intensity for the fished part of the ecosystem types (not their entire extent) is provided by estimates of mean SAR and the period between successive complete trawling of grid-cell areas (Table 2).The ecosystem types exposed to the highest mean trawling intensity included Browns Bank Rocky Shelf Edge (with fished areas covering almost its entire extent and, on average, trawled every 61 days), Southern Benguela Muddy Shelf Edge (trawled every 122 days over its entire extent), Agulhas Coarse Sediment Shelf Edge (trawled every 138 days over ~57% of its extent) and Agulhas Rocky Shelf Edge (trawled every 164 days over ~21% of its extent).8).As the majority of MPAs that overlap with trawled areas were declared in 2019, after the SAR map study period, it is important to note that the new protection is not reflected in the trawling-intensity maps and statistics.Nine ecosystem types that had the greatest proportion (≥69%) within the trawl ring-fence received protection in 2019.The Southern Benguela Muddy Shelf Edge, which was entirely within the trawl ring-fence and was exposed to relatively high trawling effort over most of its extent, now has 11% protection in the Benguela Muds MPA (Table 2; Figure 8b).Kingklip Ridge, which is near the edge of the trawl ring-fence and which was mostly exposed to relatively light trawling activity, was almost entirely protected in 2019 (Figure 8c).Browns Bank Rocky Shelf Edge, which had the highest mean SAR and >90% of its extent within the trawl ring-fence, now has 12% of its extent protected (Table 2; Figure 8d).
Altogether there are six ecosystem types that have >30% of their extent protected and two more that are >20% protected.In contrast, six of the ecosystem types that are trawled remain unprotected, of which four have notable proportions of their extent (range 39-57.3%)within the trawl ring-fence (Table 2; Table S1 in Supplementary Materials S3).

Discussion
We produced maps that represent recent (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) demersal trawling intensity at high resolution and with an intuitive and interpretable metric, with greater spatial accuracy and less error than previously accomplished in this region (e.g.Amoroso et al. 2018;Sink et al. 2019).Although satellite monitoring data are replacing the use of logbook records in mapping fisheries activity (e.g.Amoroso et al. 2018), the logbook data and resultant trawlingintensity maps in this study are expected to provide the best estimate of trawling intensity during the study period, as satellite monitoring systems have become widely implemented and their data accessible only in recent years.
The improved accuracy over similar previous work is attributed to the rigour with which errors in the underlying logbook location records were filtered or corrected.The resultant map revealed the spatial distribution of trawling activity in unprecedented detail, while annual maps were used to assess spatial and temporal changes and to identify areas of consistently high effort, suggested to inform priority fishery areas.Many of the mapped areas showed a decreasing trawling intensity over the study period, whereas few localised areas showed increased activity.
We demonstrated the utility of the trawling-intensity layer in supporting spatial management by quantifying the spatial interaction of demersal trawling activity with South Africa's benthic and bentho-pelagic marine ecosystem types and their protection levels.
Although calculated with different data and methods, the trawling-intensity index used by Wilkinson and Japp (2005) is similar to SAR in that it represents the 'area swept (km 2 ) divided by the trawlable area (km 2 )'.Their results suggested that 25.8% of trawled areas were impacted once or more per year (which they referred to as 'moderate to heavy impact').The equivalent measure from our is SAR ≥1, which relates to 54.4% of trawled areas.Similarly, their 'intensely impacted' (trawled >4 times per annum) proportion of fished area was 4.7%, whereas our equivalent threshold of SAR >4 was reached in 26% of fished grid cells.Two likely causes might explain these discrepancies in estimating the proportion of trawled areas impacted by thresholds of trawling intensity.First, the assumptions of gear width used in their calculations are smaller, on average, than ours.They assumed 110-m door widths for offshore trawl vessels and 65-m door widths for inshore vessels, whereas we used company-and vesseltype-specific widths collated by DFFE, which were 60 m for inshore vessels and ranged from 110 m to 185 m for offshore vessels.The difference in gear widths accounted for 39.3% greater swept-area estimates in our dataset than if we had used the widths assumed by Wilkinson and Japp (2005), accounting for a substantial proportion of the discrepancies and highlighting the importance of gathering accurate information on gear-width parameters.Second, another source of discrepancy might relate to the area exposed to trawling (by which the swept area is divided).Wilkinson and Japp (2005) used commercial tracks (from navigational software) and observer start and end positions to estimate the area trawled (or 'trawlable area' as they referred to it).Fished areas may change over time and will expand or contract with fishing effort (Figure 3b), which appears to have been higher at the time of the Wilkinson and Japp (2005) report than during most of our study period thereafter.The density filter applied in our study may have removed some areas of legitimate (but very low) trawling activity, and the reintroduction (upscaling) of filtered effort may have concentrated SAR in the fished areas, which may have added to the discrepancy in trawling-intensity measures of the two studies.Amoroso et al. (2018) conducted similar analyses to our study, but for 24 continental shelf regions around the globe, including two areas that were combined in our study area.Including all depths between 0 and 1 000 m as their potentially fished areas, they calculated that 9.5% and 8.6% of the South Benguela Current and East Agulhas Current shelf regions, respectively, contained 90% of trawling activity.Our equivalent estimate of 43% of the fished area and 29% of the trawling ring-fence area are much higher proportions because those areas are far smaller than the entire shelf down to 1 000-m depth.For the same reason, their mean SAR values (≤0.44) for South African shelf regions are not comparable to ours (3.15) calculated for fished areas only.The standardisation of mean trawling intensity by regional shelf reference areas by Amoroso et al. (2018) makes it difficult to compare the trawling intensities that fished seafloors were exposed to across regions.Nonetheless, maps of SAR in their supplementary materials suggest that trawling intensity was notably lower in South Africa than intensely trawled regions near mainland Europe, the United Kingdom and Ireland.
Besides its application in estimating the trawling-intensity maps, the improved trawl-location dataset provides a resource that can be used in comparison to, or together with, current and future trawl distribution data that do not suffer the same biases.The unique combination of spatial errors we encountered are not expected to be found in other similar datasets; nonetheless, some of the errors (or similar ones) are likely to have occurred elsewhere and analysts working with those data may benefit from the processes documented here.
There are a range of potential applications of high-resolution maps of fishing activities in place-based marine conservation and management.The trawlingintensity map can be used to identify core fishery areas (e.g. Figure 5) and could contribute to informing management tools, such as priority fishing areas and fisheries management areas, which might include spatial management interventions to protect important fishery areas from other activities, or to conserve strategic components of a fishery resource (Reed et al. 2020).Better maps of fishing activity will also contribute to improved assessments of their potential impacts on biodiversitysuch as in assessing the sustainability of fishing gear (Pitcher et. al. 2017), in consideration of cumulative environmental impacts in the context of strategic environmental assessments, and as pressure layers to assess the threat status of ecosystem types (e.g.Sink et al. 2019).Despite their multitude of potential uses, improved maps of fishing effort contribute one aspect of many information layers that tend to inform place-based assessments or management approaches.
The improved map of trawling intensity could contribute to planning and design for MPA expansion, through its inclusion in the cost-layer used in systematic conservation planning and in guiding the development of practical boundaries for spatial management areas.Because Table 2: Summary statistics relating to the overlap between South Africa's marine ecosystem types (see Sink et al. 2023a) and trawling areas or activity, including ecosystem type extent, area and proportion within the trawl ring-fence, area covered by trawling activity (swept area ratio, SAR value of >0), mean SAR value for the fished proportion of the ecosystem type, trawl frequency estimated as the mean number of days between successive trawling coverage of the 1-km 2 grid cells for the fished area (calculated from 1/SAR), and the area and proportion of the ecosystem type within marine protected areas (MPAs).Protection statistics are reported here following the expansion of South Africa's MPA estate in 2019, with the proportion protected prior to the expansion given in parenthesis for comparison.See Table S1 in Supplementary Materials S3 for ecosystem types prioritised for further habitat management.All area units are km 2 and proportions are %.The row sequence is sorted by decreasing proportion within the trawl ring-fence, and ecosystem types are included only if >1% of their area overlaps with the ring-fence  ), trawl ring-fence (purple outline) and protected areas (green outline) are shown on the left, while paired graphs on the right quantify the trawling intensity across the ecosystem areas, with vertical lines indicating the area outside the trawl ring-fence (purple), within protected areas (green) and 25% intervals (grey dotted lines).Note that numerous offshore protected areas were declared in 2019, after the study period, and the subsequent redistribution of trawling activity is not reflected in these graphics trawling intensity was mapped for a period (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) prior to the declaration of 20 new MPAs in 2019 (RSA 2019), the spatial patterns of trawl activity displaced by those MPAs is not captured here.Although the new MPAs were identified using systematic conservation planning in a configuration that minimised impacts on fisheries (Sink et al. 2023b), where there was no alternative to meet feature targets, some MPAs were located in areas that would have displaced trawling activity (e.g. the Benguela Muds MPA in the Southern Benguela Muddy Shelf Edge ecosystem type).The subsequent displacement of trawling patterns would be interesting to study in the context of the effects of MPA declaration on the fishery, but was beyond the scope of this work.
The 20 new MPAs declared in 2019 caused substantial progress in the protection of ecosystem types affected by trawling (Table 2).Examination of the overlap of trawling, South Africa's ecosystem types, and their protection in MPAs, can support prioritisation of ecosystem types for inclusion in future protection.For example, six ecosystem types affected by trawling still have no representation in South Africa's MPA network (Table 2).Of those, the four that have >20% of their extent within the ring-fence, namely Kingklip Koppies, Agulhas Blues, Southern Benguela Shelf Edge Mosaic and Cape Lower Canyon, are proposed as priorities for MPA expansion (Table S1 in Supplementary Material S3).There are likely to be options to protect these ecosystem types outside of the trawl ring-fence, noting that other activities and plans warrant consideration in spatial planning in this context.
Progress in management action for marine ecosystems has been an important element in maintaining eco-certification of the demersal hake trawl fishery (Sink et al. 2023b), which illustrates the benefit to the fishery of advancing spatial information, such as the trawling-intensity map showcased here and the iteratively improved marine ecosystem-type map (Sink et al. 2023a).It is important to recognise that different habitats or ecosystem types differ in their vulnerability to trawling activity (Kaiser et al. 2002;Pitcher et al. 2017) and that periods of recovery from trawl impacts will also vary across different seafloor habitats (Hiddink et al. 2017).Ecosystem types that host habitats considered to be vulnerable marine ecosystems (VMEs) (FAO 2009) may warrant higher levels of protection.However, despite their potentially misleading name, VMEs tend to most often constitute finer-scale communities within broader ecosystem types (Watling and Auster 2021), highlighting the need to map VME features at a higher resolution than their supporting ecosystems.
The SAR data layer produced in this work has been included in South Africa's first National Coastal and Marine Spatial Biodiversity Plan, which comprises a map of Critical Biodiversity Areas and Ecological Support Areas (CBA Map) and an accompanying set of sea-use guidelines to inform management within those areas (Harris et al. 2022a(Harris et al. , 2022b)).The trawling-intensity map informed the cost layer used in the Marxan software applications (Ball et al. 2009) to support systematic conservation planning from which the CBA Map was developed, helping to avoid selecting areas as biodiversity priorities that are both more valuable to the fishery and expected to be in poorer ecological condition.
The National Coastal and Marine Spatial Biodiversity Plan underpins the biodiversity sector's input into the national MSP process (RSA 2023).Consequently, areas of high priority for the demersal trawl sector have already been accounted for in identifying the biodiversity sector's priority areas, which should help reduce spatial conflict between these sectors and streamline MSP negotiations.

Future work
There are many potential applications of the trawlingintensity map in the biodiversity and fisheries sectors, as touched on above.Whether for MSP, MPA expansion, or other place-based management measures, the accuracy and resolution of the trawl map are important for effective spatial interventions and decision-making.Improvements and future updates to the trawling-intensity map and underlying data are outlined below and expanded on in Supplementary Materials S1.
Trawl gear width is a critical factor that affects calculations of swept areas (e.g.Larcombe et al. 2001).Greater certainty of trawling intensity could be achieved if reliable vesselspecific door-width records were captured in the commercial logbook data.A further improvement in the calculation of trawling intensity might be to account for varying impacts of different gear.Smaller and lighter gear used by smaller inshore vessels might, for example, have lesser seafloor impact than larger and heavier trawling gear towed by larger offshore vessels.The relative impact of different gear configurations and sizes requires investigation.
Comparison of the trawling-intensity maps from logbook records with those generated from recent satellite monitoring systems (VMS and automatic identification system [AIS]) would be a valuable next step and could inform their integration for more recent (and future) updates to the trawling-intensity map.Until the entire trawl fleet is covered by satellite monitoring systems that capture every trawl event, logbook records will continue to be a valuable data source for mapping trawling intensity.
The introduction of electronic logbooks, with automated logging of position data, would greatly benefit the quality of logbook data.Doing so would also enable the recording of tracks or multiple positions between the start and end locations, aiding the comparison to, and integration with, satellite monitoring data (VMS and AIS) for a precise and complete picture of trawling activity.

Conclusions
Although trawling is only one of several fishery sectors in South Africa, it is among the biggest in terms of social impact, geographic range and potential biodiversity impacts.As argued here, accurately mapping the fishing activities of this industry can help inform spatial decision-making that can benefit both the fishery and the environment.The development of such foundational data layers is critical to understand, manage and plan for human activities in the marine environment, even when the underlying data contain inherent errors that need to be overcome, as was the case here.Improving our ability to deliver such maps, more accurately and closer to real-time, is imperative as countries look to diversify and intensify their blue economies, accelerating trends in ocean use and impacts in the face of global climate change (Halpern et al. 2019;Jouffray et al. 2020).This is especially pertinent as countries look to achieve the Sustainable Development Goals in the post-2020 Global Biodiversity Framework (Convention on Biological Diversity 2022), as they strive for vigorous, just and sustainable economies supported by abundant biodiversity and environmental commons.

Figure 1 :
Figure 1: Map of the study area, showing start locations of 2005-2018 demersal trawl fishing records, showing isobaths (grey lines) and an outline of that part of South Africa's mainland maritime domain where demersal trawling occurs (dashed line).The two insets illustrate patterns referred to in the text at different spatial resolutions, with the transparency and size of points adjusted to highlight the relevant patterns

Figure 2 :
Figure 2: Maps showing (a) the mean trawling intensity (swept area ratio [SAR]) between 2005 and 2018, and (b) its coefficient of variation (CV) across all 14 years.Included are the trawl ring-fence (black outline) and isobaths (grey lines) in South Africa's mainland maritime domain (dashed line, grey fill).See Figure 1 for isobath labels

Figure 3 :
Figure 3: (a) Summed annual trawling activity (swept area ratio [SAR]); (b) SAR plotted against the total area that trawling occupied.Black dashed linear fit with adjusted R 2 of 0.78 is shown in panel 'b', and the grey arrows guide the sequence of years

Figure 4 :Figure 5 :Figure 6 :
Figure 4: Map showing the slope of linear fits to the annual time-series of trawling activity (swept area ratio [SAR]), indicating areas where trawling intensity has tended to increase (positive slope: brown colours) or decrease (negative slope: green colours) during the 2005-2018 study period.Included are the trawl ring-fence (black outline) and South Africa's mainland maritime domain (grey dashed line).Linear models were applied at each 1-km 2 grid cell, but the slope was plotted only when it was estimated to be different from zero (p ≤ 0.05)

Figure 7 :
Figure 7:The proportion of total area fished for (a) the range of swept area ratio (SAR) values (ranked in increasing order), indicating the area related to SAR values of <2 and <1 (dashed lines); and (b) cumulative proportion of effort (with SAR values ranked in decreasing order), indicating the area related to 90% of cumulative trawling effort (dashed line)

Figure 8 :Figure 8 :
Figure 8: Examples of ecosystem types and their overlap with the trawl ring-fence, trawling intensity and protected areas.(a) provides the spatial context of the three ecosystem types shown in (b)-(d), where the ecosystem type extent (black outline, blue fill), trawl intensity (swept area ratio [SAR] legend provided in d), trawl ring-fence (purple outline), and protected areas (green outline) are shown on the left, while paired graphs on the right quantify the trawling intensity across the ecosystem areas, with vertical lines identifying the area outside of the trawl ring-fence (purple), within protected areas (green) and 25 percent intervals (grey dotted).Note that offshore protected areas were declared in 2019, after the trawl intensity study period, and the subsequent redistribution of trawling activity is not reflected in these graphics

Table 1 :
Summary of demersal trawling records flagged for removal when missing or spurious location data could not be corrected.The 10 data-cleaning steps are outlined in the Methods section and detailed in Supplementary Materials S1.Many stations were flagged by more than one filter; hence, the last line (Total records removed) does not represent the sum of the lines above it