Bycatch species distribution from mid-water trawlers in the north-eastern Arabian sea: a step towards the implementation of marine spatial planning

ABSTRACT Spatial conservation planning to reduce the impact of fishing on ecosystems necessitates understanding of the spatial distribution of bycatch in fishing grounds. Management measures in Indian waters cannot be spatially explicit without this information. Thus, a preliminary study was conducted on the northwest coast of India to identify the most persistent bycatch areas from mid-water trawlers targeting largehead hairtails (Trichiurus lepturus). Semi-variogram modelling and kriging interpolation showed that bycatch abundance varied by species and associated habitat or depth preference. A significant positive correlation between species in bycatch abundance revealed that few share the same habitat in the region. Bycatch species were concentrated in the inshore area (<70 m depth) during the post-monsoon and summer seasons, but their densities reduced in the winter. The bycatch distribution map aid in the future implementation of spatial management measures, such as the designation of the no-fishing zone and conservation networks.


Introduction
Overfishing and fishery bycatch challenge our ability to devise sustainable management and conservation plans to maintain a healthy ecosystem, as it significantly affects the non-target resources, biodiversity, ecosystem function, and the habitat (Bijukumar & Deepthi, 2009;Dineshbabu et al., 2016;Pauly et al., 2001).The trawl fishery contributes a relatively higher proportion of bycatch (Andrew & Pepperell, 1992;Samanta et al., 2018;Tonks et al., 2008) due to the indiscriminate capture of non-target fishery resources and it is estimated to be about 27% of global bycatch discards (Eayrs, 2007;Fennessy & Groeneveld, 1997;Kelleher, 2004).The trawl fleet in the north-eastern Arabian Sea i.e. north-west Indian coast (NWCI) covers 48.81% (17,195) of total trawlers in India (MOA & CMFRI, 2012).The trawler in the region carries variants of trawl nets to target different resources.However, trawlers spend about 40-45% of fishing days in a year targeting largehead hairtail using a mid-water trawl net and this fishery contributes about 53.36% of bycatch (Azeez, Rohit, et al., 2021).Knowledge of the spatio-temporal abundance of bycatch is crucial for ecosystem-based management (EBM) measures particularly for tropical trawl fisheries such as in India, characterized by multi-species catch generally resulting in an enormous quantity of bycatch and discards challenging the sustainability of the ecosystem (Bijukumar & Deepthi, 2009;Dineshbabu et al., 2015Dineshbabu et al., , 2019)).EBM is a management strategy that considers interactions between ecosystem components, including human interactions such as fisheries (Levin et al., 2009).Spatial-temporal bycatch abundance information contributes to future EBM in the region.
Marine spatial planning (MSP) is an important tool used for the management of marine resources (Dineshbabu et al., 2019;Janßen et al., 2018;Lester et al., 2018;Meaden et al., 2016;Said & Trouillet, 2020).The spatial decision-making process in the coastal and offshore waters helps to enhance fisheries and aquaculture sustainability in addition to traditional fisheries management measures such as bycatch excluder devices, mesh size, and effort control (Alverson et al., 1994;Dineshbabu et al., 2019;FAO, 2014).Spatial mapping of marine resources will be helpful for stakeholders and managers in the decision of where to require more attention i.e. fishery and bycatch hotspots (Azeez et al., 2016;Colloca et al 2015; Koya, Rrohit, et al., 2018;Morfin et al., 2012).Spatio-temporal abundance of the major resources in the fishery mapped for development of spatial conservation planning in the various regions such as in the north-western Mediterranean Sea (Morfin et al., 2012;Saraux et al., 2014), European Union Mediterranean waters (Colloca et al., 2015), Western Australian waters (Aidoo et al., 2015), south-eastern Arabian Sea (Dineshbabu et al., 2015).However, such information is lacking in the north-eastern Arabian Sea, limiting the application of spatially explicit management measures.We proposed a geostatistical approach to mapping spatio-temporal distributions of the key bycatch species in the mid-water trawlers operating in the NWCI.

Study area
The NWCI faces the north-eastern Arabian sea (Latitude: 15.5-23.3°N,Longitude: 66.5-73.5°E; Figure 1) including two coastal provinces of India, Gujarat and Maharashtra, which together accounted for 33% (2.96 lakh km 2 ) of the continental shelf area and 28.91% (2,347 km) of the coastline of India (MOA & CMFRI, 2012).A wider continental shelf in this region provided extensive fishing grounds and the region contributed 32.1% (1.14 million t) to the total marine fish landing in India in 2018 (CMFRI, 2019).

Fishery data
Fishery data were collected from the logbook of three multiday mid-water trawlers operating along NWCI from August 2017 to December 2019.The multiday mid-water trawlers operated from Veraval, a major harbor in the region were selected for geo-spatial data collection of the fishery.Azeez, Rohit, et al., (2021) revealed mid-water trawler operations and gear specifications used for targeting largehead hairtail in the NWCI.The total weight of each species in a haul was standardized using catch per unit effort (CPUE) in kilogram per hour (kg h -1 ) of the fishing operation.CPUE for each species in the haul is derived from Equation (1).
where CPUE sijt is kilogram of a species s caught per hour (kg h -1 ) in i and j location at t time; w s is weight of the species; h j is time haul and h i is a time of the shoot.
Fishery data with CPUE of key species were highly skewed with few extremely large values and high proportion of zero and small values.Hence, Box-Cox transformation applied to correct frequency distribution of CPUE as γ = log (χ+c), where χ is CPUE and c is a positive constant (c we set to the smallest observed CPUE).Fishery data were grouped into three based on the fishing season viz post-monsoon (August-November; n = 347), winter (December-February; n = 274), and summer (March-May; n = 213) (Azeez et al., 2016;Koya, Rrohit, et al., 2018;Solanki et al., 2016).June and July months, the peak South-West monsoon period is a legally enforced fishing holiday in the region and hence there is a data gap.The catch of each species was averaged into 50 km × 50 km grid block catch for the different seasons to reduce the influence of excess fishing effort; due to this it can be assumed that the fish caught in the gear are unbiased indicators of their spatial abundance pattern.Such unbiased data can improve the overall performance of the model (Brodie et al., 2015;Pennino et al., 2016).CPUE of species in each grid block retrieved using Equation (2).
where CPUE spg is the catch rate of a species sp in a g grid block, CPUE spt is CPUE of species sp in total t, N g is the number of trawling observed in a grid block g.

Key species in the fishery
Although more than 100 different species of teleosts, cephalopods, crustaceans, and elasmobranchs form bycatch in the mid-water trawlers targeting largehead hairtail in the NWCI (Azeez, Rohit, et al., 2021), many are very low in abundance or are even rare.Therefore, the present study focuses on the target species i.e. largehead hairtail and the bycatch species that contribute 50% of the total bycatch (in weight).This selection led to retaining the following nine species i.e.Trichiurus lepturus, Uroteuthis duvaucelii, Sepia elliptica, Nemipterus japonicus, Otolithes cuvieri, S. pharaonis, Priacanthus hamrur, Sepiella inermis, and Saurida tumbil (Supplementary Table S1).These nine species together contribute about 73.32% to the total catch (in weight) (Azeez, Rohit, et al., 2021), hence are called key species of mid-water trawl fishery in the NWCI.

Analysis of spatial structure
Fishes tend to aggregate in different areas of a region depending on the biotic and abiotic factors (preys, environmental variables, mating, etc), which are also spatially structured (Badenhorst & Smale, 1991;Chang et al., 2012;Das et al., 2017;Lan et al., 2017).
The spatial structure has a value of spatial autocorrelation at different lags (distance intervals), and such characteristics of spatial structure in fish distribution affect the catch rate in the fishery.Therefore, the geostatistical technique of semi-variogram is used for analyzing the spatial pattern (spatial autocorrelation) of key species in the fishery (Matheron, 1963).Semi-variogram is defined as the average squared difference (h) between points.Semi-variogram model was fitted for all the key species in three seasons i.e. post-monsoon, winter, and summer.Empirical semi-variogram functions were calculated and theoretical spherical semivariogram models fitted by the weighted least-squares method using Equations ( 3) and ( 4), respectively for the key species of the mid-water fishery.
where γ (h) is the semi-variance function, N(h) is the set of all pair-wise Euclidian distances h = x−y, |N (h)| is the number of discrete pairs in N(h), and Z x and Z y are data values at location x and y, respectively (Diggle & Ribeiro Jr, 2007;Goovaerts, 1997).
where C 0 is the nugget effect, C is the dispersion of variance, C 0 +C estimates the variance of a random process which is known as the sill (Oliver & Webster, 2014), and k is the range.The spherical model is commonly used in the geostatistical analysis of the biological population (Freire et al., 1992), due to its structure of linear increase at the origin and followed by stabilization to an asymptote, which corresponds with a spatial variation that is often observed in nature (Lloyd, 2006).The consistency of semi-variograms (spherical model) was evaluated using the jack-knifing test.The jack-knife test has been increasingly used to examine the accuracy of model performance (Rehman et al., 2013).The estimate is considered adequate when the mean (µ jk ) and variance (σ jk ) of the reduced error are close to zero and one, respectively (Vieira et al., 2010).Geostatistical analyses were performed using the 'geoR' and 'RGeostats' packages (Renard et al., 2021;Ribeiro Jr et al., 2020), in R version 3.6.1 software (R Core Team, 2019).
The degree of spatial dependence (ISD) is used for analyzing spatial structure and variability.ISD is calculated from the ratio in percentage between the nugget effect (C 0 ) and the level given by the sum of the nugget effect and structured variation (C) using Equation (5).

Spatial distribution of the key species
The center of gravity and inertia has been used to characterize the spatial distribution of the key species and to investigate seasonal variations (Saraux et al., 2014;Woillez et al., 2007).The center of gravity denotes the mean location of the population and the mean location of an individual fish taken at random in the field, whereas inertia describes the dispersion of the population around its center of gravity.Seasonal maps of log-CPUE were produced for the key species by kriging interpolation on a 0.5 km 2 mesh grid over the study area using seasonal semi-variogram.Subsequently, an annual average map for each species was created by averaging seasonal maps.The kriging interpolation was carried out using the 'geoR' package in R (R Core Team, 2019 ;Ribeiro Jr et al., 2020).The thematic map of the coastline was downloaded from an open-source (https://thematicmapping.org/downloads/world_borders.php) and maps in the study area were projected to Universal Transverse Mercator (UTM) zone 42 N with geodetic datum of WGS 1984 using the 'sf' package (Pebesma, 2018) in R.
The Empirical Orthogonal Functions (EOF) were used to quantify the temporal stability/variability of spatial distribution (Lorenz, 1956).EOF analysis is similar to the Principal Component Analysis applied to the spatio-temporal data.Seasonal kriging maps of each species with time (season) as a descriptor and space (pixel) as objects are used for EOF analysis to compare the seasonal spatial distributions independently of their level of abundance.The first axis of EOF (eigenvector) is the linear combination of the seasons which maximizes the variance of the spatial distributions and its percentage of variance is an indicator of the temporal stability/variability of the spatial distribution.A higher temporal stability of the spatial distribution accounts for a greater the percentage of variance.EOF performed using the 'STRbook' package (Wikle et al., 2019) in R.

Spatial structure
The abundance of largehead hairtail had a weak significant correlation with key bycatch species except U. duvaucelii (r = 0.35) in post-monsoon, U. duvaucelii (r = 0.62), S. elliptica (r = 0.31), O. cuvieri (r = 0.43) and S. tumbil (r = 0.32) in winter and U. duvaucelii (r = 0.31), and S. tumbil (r = 0.38) in summer (Figure 2).Seasonal empirical variograms on log-CPUE values show spatial structures as semivariograms increased linearly with distance for all the species, the variograms were modeled by a spherical function that stabilized around a sill or had an additional linear structure starting from a distance range (Figure 3).The seasonal variations in the spatial structure of the fishes revealed the seasonal fluctuations in the catch rate of the species.ISD was found moderate for all the key species throughout the season except T. lepturus in post-monsoon and S. elliptica and S. inermis in summer had strong ISD, whereas N. japonicus, P. hamrur and S. tumbil during summer had weak ISD (Table 1).

Spatial distribution and mapping
Center of gravity and inertia showed seasonally shifted latitudes within the key species except for T. lepturus and U. duvaucelii (Figure 4) confirming the tendency of spatio-temporal variation of key species catch rate.Figures 5-7 show the peak abundance were spatially varied between season for T. lepturus, U. duvaucelii and S. elliptica, respectively, as examples (see Supplementary Figure S1, S2, S3, S4, S5, S6 for other species and Main Map).The log-CPUE of T. lepturus and U. duvaucelii appeared high in the northern region of the study area during the postmonsoon.In contrast to this, a higher abundance was found off Maharashtra coast for the remaining resources.Furthermore, N. japonicus and S. tumbil exclusively were inshore waters (<50 m depth).The peak abundance of T. lepturus extended south up to 19°N during winter, whereas S. elliptica, N. japonicus, O. cuvieri and S. inermis were shifted to offshore waters (>50 m depth) of the northern part (>21°N).A lower catch rate appeared for the majority of the species in inshore waters of the southern region during winter.U. duvaucelii, S. pharaonis, and P. hamrur displayed a patchy distribution in the offshore waters.Spatial distribution of catch rate was homogenous in the study area except for deeper areas off Gujarat (>100 m depth) between 21°N and 21.5°N for U. duvaucelii, S. elliptica, O. cuvieri, S. pharaonis and S. inermis during summer.In contrast to this, high bycatch rates appeared for P. hamrur in the deeper area off Gujarat.EOF was applied on the 3 seasonal maps of each species confirming the high temporal variation in the spatial distributions.The first axis of EOF explained the percentage of variance that varied among species from 59% for P. hamrur and 77% for T. lepturus.Figure 8 shows a projection on the first axis of EOF for each species.

Discussion
Fishes are continuously moving animals from one place to another during the various season for feeding  and reproduction, and their distance of vary from species to species (Griffiths et al., 2010;Binder et al., 2011).The spatio-temporal structure we modeled in our analyses was likely a result of a complex interaction between biotic and abiotic processes in NWCI, including predator-prey interactions and environmental forcing.Cyclonic current from the Somalian coast and dry cool winds from the land area during south-west and north-east monsoon, respectively affect the physicochemical and biological productivity of the NWCI (Solanki et al., 2016); thus affect the spatio-temporal variations in catch and bycatch rates in the mid-water trawlers operating in the region (Azeez, Rohit, et al., 2021).Largehead hairtail (T.lepturus) is known to have unique spatio-temporal variations in their abundance for the feeding and cycle (Azeez et al., 2016;Ghosh et al., 2014;James et al., 1986;Narasimham, 1972).Furthermore, environmental factors also influence the abundance and movement of the species (Azeez, Raman, et al., 2021;Martins & Haimovici, 1997).
The key species exhibited a significantly positive correlation coefficient with the largehead hairtail indicating that they share similar habitats.Like largehead hairtail, the loliginids show spatio-temporal movements linked to their spawning and feeding choices (Martins et al., 2006;Martins & Perez, 2007;Roberts & Sauer, 1994).They aggregate in large numbers at depths between 30 and 170 m during the spawning season, and higher abundances are found in inshore waters (Silas et al., 1982).In NWCI, a major spawning peak for the U. duvaucelii during the post-monsoon (Mohamed, 1993;Silas et al., 1985).These spatiotemporal movements of U. duvaucelii are reflected in the present study.In contrast to this, cuttlefishes (S. pharaonis, S. elliptica, and S. inermis) adults aggregate in coastal waters for spawning, and later juveniles migrate from nursery grounds to offshore deeper waters (Denis & Robin, 2001;Dunn, 1999;Wang et al., 2003).The major spawning season of cuttlefishes is during post-monsoon and summer along the Indian coast (Neethiselvan et al., 2002;Sundaram, 2014;Sundaram & Khan, 2011).Introduction of fish trawls the Indian coast leads to spatial expansion of the fishing grounds and emerged new fishery resources in recent decades.For instance, P. hamrur was a non-conventional fishery resource and became a major component of the trawl fishery of the region as species preferred relatively deeper waters (Kizhakudan & Zala, 2006;Seetha et al., 2018;Sivakami et al., 2003).This has been corroborated by the findings of the present study that seasonal maps of P. hamrur abundance were higher in offshore waters (>50 m depth) except during postmonsoon.Similarly, spatial distribution of threadfin breams, lizardfishes, and bull's eye reported that abundance was high in deeper waters (>100 m) in the region (Selvaraj et al., 2007;Vivekanandan, 1990).Our findings agreed with Bapat et al., (1982) andVagh, (2018), that the abundance of N. japonicus is higher along the Maharashtra coast.The distribution of S. tumbil was similar to N. japonicus and a high positive correlation coefficient exists between the spatial abundance of these two resources indicating that these resources share identical habitats.
Spatial-temporal distribution of fishes is directly related to their prey availability (Csepp et al., 2011;Li et al., 2020).For instance, the most preferred prey items of largehead hairtail were cephalopod, sciaenid, threadfin bream, lizardfish, etc. (Ghosh et al., 2014;Koya, Vase, et al., 2018;Rohit et al., 2015) which are the key bycatch species in the mid-water trawl targets largehead hairtail in NWCI.Largehead hairtail is known for its association with other fishes for trophic competition (Azeez, Rohit, et al., 2021;Martins & Haimovici, 1997).Hence, a high proportion of bycatch and discards in the fishery, mainly trawl operated in the inshore waters (Azeez, Rohit, et al., 2021), because inshore coastal waters are regarded as the nursery grounds of fishes (Camp et al., 2011;Sheaves et al., 2014).Mid-water trawler net is characterized by a wide vertical opening to cover a wider water column unlike shrimp trawler net, and when operated in the inshore waters resulting high proportion of juvenile commercial fishes and non-target resources (Azeez, Rohit, et al., 2021).Moreover, coastal upwelling and primary productivity during monsoon influence the fish distribution in the region (George et al., 2012;Solanki et al., 2016) which triggers the movement of fishes to coastal waters for feeding during post-monsoon; resulting in larger scale capture by trawlers during this season (Ghosh et al., 2009;Murty et al., 2003).Therefore, the map depicts spatial abundance of key bycatch species that were high along the inshore waters during post-monsoon.Hence, spatial demarcation of fishing grounds is necessary for the coastal waters to achieve sustainability of the fishery, besides traditional management measures such as juvenile excluder devices and mesh size regulation.Information on the spatial-temporal aggregation of fish is crucial for the spatial management of the fishery in the coastal waters and EBM (Dineshbabu et al., 2019;Koya, Rohit, et al., 2018).Therefore, the present finding gives support fishery managers and policymakers for future development of MSP as a management tool for highly crowded fisheries such as in NWCI.Further development of maps for the spatial distribution of adults versus those of juveniles and modeling the environmental suitability of fishes using vector autoregressive spatio-temporal model (VAST) and other machine learning may also improve the formance of the MSP as well as a marine protected area.

Conclusion
Understanding the spatio-temporal distribution of marine fishery resources is important for managers and policymakers.Tropical coastal waters such as India have high species diversity with multiple species being involved in the fishery and it poses a major challenge for resource managers.The creation of a geospatial data layers of the resources can form a key tool to track the problems associated with bycatch in the fishery for resource managers and policymakers.The present study shows spatio-temporal abundance pattern of largehead hairtail and key bycatch species can be attributed to the unique feeding and reproduction biology of fishes.Therefore, monthly or seasonal mapping of fishery resources can contribute significantly to the development of MSP and spatial management strategies.spatial regulations of fishing operations such as no-fishing zone be designed based on such explicit information on the respective fishery.However, future spatial regulations need to be evaluated for their socio-economic implications to minimize the conflict with resource users.Such spatially regulated fishery can reduce the quantity of bycatch, improve capture of high valued target fishes and economic efficiency of fishing operation that makes fishery more sustainable.Software R version 3.6.1 software (R Core Team, 2019) for geostatistical analysis and ESRI ArcGIS 10.8 was used for generation of final maps.

Figure 1 .
Figure 1.Study area on the north-west coast of India.The shaded area indicates the number of fishing trips observed in the 50 km × 50 km grid in the study area.Study area map projected to UTM, Zone 42 N.

Figure 2 .
Figure 2. Pearson correlation plot of key bycatch species from mid-water trawl operating in NWCI during (A) post-monsoon, (B) winter, and (C) summer.

Figure 3 .
Figure 3. Seasonal semi-variogram modeled for key bycatch species.The red, blue and green lines indicate post-monsoon, winter and summer, respectively.

Figure 4 .
Figure 4. Seasonal center of gravity and inertia axis of key species.The red, blue and green lines indicate post-monsoon, winter and summer, respectively.

Figure 8 .
Figure 8. Projection on the first axis of Empirical Orthogonal Function performed on kriging maps of log-CPUE of key species. ., CONTACT P. Abdul Azeez abdul.p@icar.gov.in;azeez.cr7@gmail.comICAR-Central Marine Fisheries Research Institute, Kochi 682018, Kerala, India; ICAR-Central Institute of Fisheries Education, Mumbai 400061, Maharashtra, India This article has been corrected with minor changes.These changes do not impact the academic content of the article.Supplemental map for this article is available online at https://doi.org/10.1080/17445647.2023.2192371.

Table 1 .
Geostatistical parameters from semi-variogram model and their corresponding spatial dependence for key species from mid-water trawl operated in NWCI.Index of Spatial Dependence, µ jk = average reduced error and σ jk = reduced error variance.