Evolution of the Food Web in Bandon Bay, the Gulf of Thailand: Ten Years of the Blue Swimming Crab Stocking Programme

The ecosystem of Bandon Bay, in the Gulf of Thailand (GoT), has been impacted since 2007 by the continued stocking of larval blue swimming crab Portunus pelagicus, also called a crab bank. In this study, the food web in the Bay was modelled using Ecopath software to compare the trophic status, interaction and energy flow among the components in the system in 2007 and 2016 (i.e., before and 10 years after the crab bank intervention). The models were based on data collected from trawling. Twenty fish and shellfish components were used in the 2007 model, while 22 were used in the 2016 model. A significant increase in biomass was found in blue swimming crab, but biomass declined for other demersal fishes, cephalopods, and Penaeid shrimps. The production/biomass ratios of most components were higher in 2016 but the consumption/biomass ratios were relatively unchanged. The ecotrophic efficiency indicated that shellfishes were more exploited than fishes. Changes in most of the ecological indices revealed higher maturity and stability after 10 years of crab bank operation. The mixed trophic impact indicated bottom-up regulation, and that the increase of blue swimming crab negatively impacted only Mantis shrimp. Overall, the results indicate positive impacts of the crab bank intervention.


INTRODUCTION
The Gulf of Thailand (GoT) is among the most productive large marine ecosystems; the marine capture fisheries within Thai territory of the GoT contribute over 65% of the country's total marine production (about 2.5 × 10 6 tonnes) each year (Lymer *Corresponding author: tuantong.j@ubu.ac.th One of the most serious concerns for a stocking programme, including the crab bank, is whether this activity causes changes in the abundance of other species in the system, which could consequently lead to imbalance of populations and possibly result in the loss of other ecosystem values and services (Caddy & Defeo 2003;Molony et al. 2003;Bell et al. 2006). This imbalance is mainly from two causes: competition and predation. Competition for food resources occurs both at the intraspecific level, due to increased abundance of individuals by the addition of hatchery-reared seeds, and at the interspecific level, due to competition between hatchery reared seeds and other species with similar ecological requirements and potentially leading to a reduction in abundance of competing species and prey species (Molony et al. 2003). Predation can occur either by or to the stocked species, which may result in trophic cascades, or community-level cascades (Polis et al. 2000) that impact at least three trophic levels and can extend to any multilink linear food-web interaction (Caddy & Defeo 2003). Moreover, exceeding the carrying capacity of the ecosystem due to continued stocking is also considered a cause of imbalance (Blaxter 2000;Molony et al. 2003). Therefore, quantification of the impacts of stocking programmes, such as the crab bank, on the ecosystem is an important step in determining the appropriateness of particular management actions (Fayram et al. 2006;Khan et al. 2015).
Understanding of food web structure and ecosystem dynamics is important for determining the interactions in an ecosystem and useful to many ecological studies (Khan et al. 2015). Several mass-balance models have been applied for the purpose of understanding ecosystem processes and how they govern the living components in the system. Among the mass-balance models, Ecopath (Polovina 1984) is the most popular and widely applied for estimating the biomass budget for each component in the ecosystem, together with their mortality, diet and energetics parameters. Ecopath partitions the ecosystem into boxes representing a component, i.e., a species or a group of species that have similar life history. The software analyses interactions among components as well as provides quantitative descriptions of the structure of food webs of the system. In doing so, Ecopath works under the assumption that the ecosystem under consideration is at equilibrium, i.e., inputs to a component should equal outputs for the period being considered (Polovina 1984;Christensen et al. 2005). As a software for balancing steady-state model, it allows the user to make a comparative study between two periods of interest, in particular before and after intervention by human activities such as regulation measures; fisheries actions, damming, species introduction as well as stocking programmes (Christensen et al. 2005;Fayram et al. 2006;Khan et al. 2015).
This study, therefore, aims to describe two different situations of the Bandon Bay ecosystem in 2007 and 2016, and investigate the evolution of the ecosystem through its food web structure and ecosystem functioning in response to the stocking of P. pelagicus through the crab bank programme. It is worth noting that the year 2007 was the first year of the "crab bank" campaign in Surat Thani Province; later these crab banks were implemented more intensively along the coast of Bandon Bay (Sawusdee 2010). The study was done using the Ecopath with Ecosim (EwE) software version 6.2 (freely available at http://www.Ecopath.org; Christensen et al. 2005). The results can be further applied for policy development on the sustainable use of resources in Bandon Bay or for deriving management strategies for blue swimming crab fishing grounds elsewhere.

The Study Area
Bandon Bay (9°12' N; 99°40' E), located in southern Thailand (Fig. 1), is the largest estuarine (ca 1,070 km 2 ) and mangrove inlet on the east coast of Thailand, and empties into the GoT. This bay serves as a crucial nursery and feeding ground of many brackish and marine species and is considered a textbook example of excessive utilisation of coastal resources (Jarernpornnipat et al. 2003). Surface water currents in the bay show two significantly different patterns, according to season: counterclockwise circular patterns during the dry season, from January to March; and flowing southwards during the rainy season, from April to December (Wattayakorn et al. 1999). The coastal area is gradually sloped, and the average water level in the bay is 2.9 m, fluctuating from less than 1 m to 5 m (Wattayakorn et al. 1999;Jarernpornnipat et al. 2003).

The Ecopath Model
Since the first introduction of the Ecopath model in 1984 in French Frigate Shoals (Polovina 1984), this model has been widely used to describe the trophic interactions and mass balance in aquatic ecosystems. It uses the Ecopath with Ecosim (EwE) software, and the model has been progressively improved, both in terms of software and techniques, by the University of British Columbia's Fishery Centre (Christensen et al. 2008;Heymans et al. 2016). Details of Ecopath and instructions for constructing models with it can be obtained from the website, http://www.Ecopath.org, or by viewing examples of over 400 models published in various scientific journals (Colléter et al. 2015;Heymans et al. 2016). In brief, for the Ecopath model, it is assumed that the ecosystem is in steady-state for each component, i.e., inputs equal outputs, and the basic mass-balance concept (Christensen et al. 2005) can be described as: Production = catches + predation mortality + biomass accumulation + net migration + other mortalities (1) or written as a linear equation as: where, for any component (i), P i is the total production rate; Y i is the total fishery catch rate; M2 i is the total predation rate; B i is the biomass; E i is the net migration rate (i.e., emigration -immigration); BA i is the biomass accumulation rate; M0 i = P i × (1 -EE i ) is the other mortality rate, and EE i is the ecotrophic efficiency (i.e., the fraction of the production that is utilised within the ecosystem by predators or exported or removed by fishery).
To construct the ECOPATH, the model is expressed in terms of utilisation of production of each component in the ecosystem at an arbitrary time period, and Equation (2) can be re-expressed as: where (P/B) i is the production/biomass ratio; B j is the biomass of predator j; (Q/B) j is the relative food consumption of j; DC ij is the fraction of prey i in the diet of predator j; EX i is the export from the ecosystem, mostly through fisheries. From Equation (3), four parameters, namely B i , (P/B) i , EE i and (Q/B) j , as well as diet composition of each component are required as inputs to construct the ECOPATH. At least 3 out of 4 parameters must be input to the model for each component, and then n linear equations are created for n components and solved for the remaining parameter (Christensen et al. 2005).

Model components
Details of component in the Ecopath analysis of the Bandon Bay ecosystem are in Table 1. Finally, there were 20 fish and shellfish components (i.e., species/group of species) used for constructing the Ecopath model of Bandon Bay in 2007, and 22 components for the 2016 model (Tables 2 and 3). These components represent the catch composition from trawl surveys by the research vessel of the Chumphon Marine Fisheries Research and Development Centre within the Bandon Bay area, which were six survey-cruises in 2007 and 10 in 2016. (Continued on next page)   Note: P/B is production/biomass ratio, Q/B is consumption/biomass ratio, EE is ecotrophic efficiency, and P/Q is production/consumption ratio or gross efficiency (GE)  Notes: P/B is production/biomass ratio, Q/B is consumption/biomass ratio, EE is ecotrophic efficiency and P/Q is production/consumption ratio or gross efficiency (GE)

Model inputs
Input parameters for the basic estimation in the Ecopath model are shown in Tables 2 and 3 (Sparre & Venema 1992) as: where CpUE is the average catch per unit effort of each component; a is the area swept by the trawl per hour (0.09029 km 2 ); X 1 is the proportion of fish in the path of the gear retained by the net (0.5) and A is the total area of Bandon Bay (480 km 2 ).

Production/Biomass ratio (P/B):
The P/B ratio was estimated through use of the instantaneous rate of total mortality (Z, year -1 ) as described by Allen (1971). During the surveys, catch of each species was sampled and lengths of individuals were measured. Thus Z was estimated by Beverton and Holt (1957) as: where L ∞ is the asymptotic length (cm), K is the curvature parameter of the von Bertalanffy's growth function, L is the mean length in the population (cm), and L' represents the mean length at entry into the fishery (cm).
3. Relative food consumption (Q/B): The Q/B ratio was estimated from the empirical relationship proposed by Palomares and Pauly (1989) as: where W ∞ is the asymptotic weight (g), T' is the mean temperature of Bandon Bay at 29°C (expressed by T' = 1000/K (K =°C + 273.15), A is the aspect ratio (A = H 2 /S; H is the height of caudal fin and S is the surface area) for a given fish, h is a dummy variable expressing food type (1 for herbivores, and 0 for detritivores and carnivores), and d is a dummy variable also expressing food type (1 for detritivores, and 0 for herbivores and carnivores). The aspect ratio of each fish as well as Q/Bs for the shellfishes were derived from Vibunpant et al. (2003).

Diet composition: the input on diet composition of each component was derived from relevant scientific reports on fish stomach contents in Bandon
Bay and adjacent areas by DoF marine fishery scientists (Tables 4 and 5).
5. Inputs of non-fish and non-shellfish components: Biomass, P/B and Q/B of these components (benthos, zooplankton, phytoplankton and detritus) were derived from relevant literature (Supongpan et al. 2005a;2005b;Sawusdee et al. 2009;Premcharoen 2012) and were assumed constant during the studied periods.

Model balancing
After input of all required parameters (biomass, P/B and Q/B, and diet composition data) into the model, a mass-balance was performed by modifying the entries until input and output were equal for each component (Webber et al. 2015). The criterion used for balancing the model was that the EE values for each component must be less than 1.0. If an EE value is more than 1, it indicates that predation on the component is greater than its production. Moreover, the gross efficiency (GE), i.e., food conversion efficiency, of each component in the system should range between 0.1 and 0.3 (Christensen et al. 2005). Thus, to meet the criteria for balancing the model, subtle adjustments were made for diet composition.

RESULTS
Components (species/ species groups) in the models for 2007 and 2016 were similar except for stingrays, i.e., Family Dasyatidae, which were not recorded in the 2007 surveys. Some species were added to other component groups because their biomass was minimal during the two surveys (Table 1). Differences in biomass among the fishery resource components of Bandon Bay were observed after the ten-year interval. Most of the fish groups showed an increase in biomass, including the blue swimming crab P. pelagicus. A significant increase in biomass of blue swimming crab was observed despite high fishing pressure on this species, which was comparable between the two periods, and this may imply the success of the stocking programme (Table 2). On the other hand, three components showed significant decreases in biomass: other demersal fishes, cephalopods and Peaneid shrimps. The P/B values (estimated through Z-value) of most components in the 2016 model were a bit higher than 2007 models, except for Lagocephalus spp., pony fish, scads and Upeneus spp. This is due to the smaller average size of the samples in 2016. Meanwhile Q/B values were set as constant in both models, i.e., assumes no difference in feeding rate of individual components. The trophic level (TL) of all components showed non-substantial changes, i.e., the difference in TL of each component between the two periods was less than 0.5, which implied their feeding plasticity. The TL of the blue swimming crab was 2.75 in 2007 and 2.54 in 2016.
The basic inputs and estimated parameters (EE and GE, as presented by P/Q) from the Ecopath model for Bandon Bay for 2007 and 2016 are presented in Table 2; the diet composition of each component is presented in Table 3. The EE values of all components were less than 1, and the GE values ranged between 0.1 and 0.3, meeting the requirements of a balanced model (Christensen et al. 2005) for both Ecopath models. The EE values for the shellfish components (> 0.5) were higher than the fish components (< 0.5), indicating that shellfish species were more heavily exploited than fishes in Bandon Bay. The blue swimming crab was among the components that were highly utilised both from within (through predation) and outside (through fisheries) the system, since its EE was close to 122 1.0. The EE values of the fish components were relatively low, indicating they were less predated on by the other components in the system. In terms of GE, i.e., food conversion efficiency, the value of 0.25 for the blue swimming crab indicated that consumption was four times higher than production. The balance network analysis (Fig. 2) shows the interactions and energy flows among components in the system. It is clear from this that the blue swimming crab mostly depended on the detrital-based food chain, i.e., the trophic interactions among recycling organic matter, detritus, predators on detritus (i.e., zoobenthos and zooplankton), and finally its predators. Considering the system statistic estimates (see Table 6) for the Bandon Bay models, most of the ecological indices showed higher maturity and stability after 10 years of stocking blue swimming crabs. The throughput value of the 2007 phase (15071.19 t km -2 y -1 ) is a bit larger than the 2016 phase (11304.34 t km -2 y -1 ), which could be due to the fisheries in the Bandon Bay, which are mostly artisanal, except for the commercial blue swimming crab fishery. The Bandon Bay ecosystem became more mature from 2007 to 2016, as indicated by the total primary production per total respiration (TPP/TR), which was 2.06 in 2007 and 1.30 in 2016. The development of the Bandon ecosystem towards maturity during the 10 years of crab stocking also was reflected by higher values of system omnivory index (SOI), total number of pathways, and % ascendency in 2016 than in 2007. The higher total number of pathways and mean length of pathways in 2016 implied that the food web in the Bandon Bay ecosystem became more resistant to perturbation. The mixed trophic impact (Fig. 3) describes the impact to all components in the system when the abundance of any impacting groups slightly increases, i.e., 10%.
Increase of natural food sources (detritus, zooplankton, zoobenthos, phytoplankton and plants) showed positive impact on most of the remaining components, indicating bottom-up regulation in the Bandon Bay ecosystem. Increase in abundance of carnivorous fish (i.e., TL > 3), resulted in negative impact on most fish groups within this ecosystem as well as themselves, i.e., by cannibalism. The mixed trophic impacts (Fig. 3) clearly indicated that the increase in abundance of the blue swimming crab resulted in negative impact only to mantis shrimp by interspecific concentration, i.e., niche overlap.

DISCUSSION
Applying the Ecopath model allows us to describe the trophic interactions and balance the biomass and annual production of key components in the Bandon Bay ecosystem before (2007) and after initiation of the crab bank programme (2016). The focus of the study was the blue swimming crab, which was continuously released into the studied area since 2010. Comparing the two Ecopath models showed differences in the food web structure and ecosystem properties in the Bandon Bay ecosystem that occurred during the 10 year interval. The major changes in the ecosystem properties of the bay were observed in the summary statistics attributes (Table 4), which showed improvement of ecosystem health. Although this improvement was certainly due to multiple causes, it may also be concluded that there was no negative effect to the ecosystem from the crab bank practice. It can be said that the Bandon Bay ecosystem became more mature, since TPP/TR in a mature ecosystem should be equal (Odum 1969); in this study the ratio decreased from 2.06 in 2007 to 1.30 in 2016. The connectivity index (CI) and (SOI) are correlated with system maturity because food chains generally change from linear to web-like as a system matures (Odum 1969;Khan et al. 2015). In this study, although CI did not change, SOI was higher in 2016, indicating the more web-like system. All flows and biomasses in the ECOPATH model can be shown in a single flow diagram as in Fig. 2, in which the size of the circles is proportional to biomass for each component and position on the y-axis represents trophic level. Also, according to Odum (1969), most components depended more on the detrital pathway, and this was apparent in 2016.
The EE values indicated that most components were substantially utilised, both from predation and exploitation in the system. It seems that the EE of most fish components in Bandon Bay were relatively low when compared to the whole GoT, for which values are always > 0.90 (Vibunpant et al. 2003;Supongpan et al. 2005a;2005b). This could be explained by the bay per se acting as a nursery ground, and the fishing area and gears used are limited, mostly for artisinal fisheries (Jarernpornnipat et al. 2003;Sawusdee 2010). Moreover, the main fishery targets in the bay are shellfishes, i.e., squids, mantis shrimp, shrimps and blue swimming crab (Sawusdee 2010;Niumnuch & Purisumpun 2011), which also had higher EE than the fish components. The higher EE values for natural food sources (detritus, zooplankton, zoobenthos, phytoplankton and plant) indicated that they were nearly fully utilised by organisms in higher trophic levels (Khan et al. 2015); in particular, phytoplankton seems to be the base food source in the Bandon Bay ecosystem (Lursinsap 1982). The substantial increase in biomass of the blue swimming crab in 2016 likely led to a consequent increase in EE of the detritus and benthos, because of the crab's bottom-feeding behaviour (Caddy & Defeo 2003). Duldic et al. (1997) mentioned that coastal areas are usually comprised of low trophic level species with high ecological efficiency and productivity, which support the carnivores within or beyond the system. The majority of the biomass in 2007 and 2016 came from components with TL between 2 and 3. There was little variation in TL for these components in both periods, indicating that although they feed mainly on their preferred diet items, they have the capability for feeding plasticity (Panikkar & Khan 2008;Duan et al. 2009). Meanwhile, the decrease in TL of the blue swimming crab in 2016 may have been caused by intra-specific competition, whereby the increased abundance through stocking caused individuals to feed more often on detritus instead of the common prey, i.e. zoobenthos and zooplankton (Kunsook et al. 2014). The mixed trophic impact showed the characteristics of bottom-up control in the Bandon Bay ecosystem, in which changes in abundance of components with TL = 1 had positive impacts on most of the other components at higher trophic levels, and these impacts dominated ecosystem processes (Dyer & Letourneau 2003;Chassot et al. 2005). The possibility of a trophic cascade in Bandon Bay can also be considered. High fishing pressure on the shellfish components would result in a shift of diets of high-TL (i.e., > 3) components.
Jutagate and Sawusdee (2022) showed that the bottom-set gillnets and collapsible crab traps, the main fishing gears in blue swimming crab fisheries of Bandon Bay, are both focused exclusively on crabs, and that the crabs contributed over 50% of the index of relative importance of the catches. Considering the results of mixed trophic impacts, this implies that if there was excessive effort from both fishing gears, imbalance in the ecosystem would occur in the system. Some fishes such as ponyfish and fishes in Family Sciaenidae would be impacted by losing their preferred food source (i.e., blue swimming crab), and predate more on other invertebrates instead. Moreover, other species that were caught substatntially in either gear type, for example, horseshoe crab in gillnets and puffer fish and Murex snail in traps, would be reduced and consequently affect their prey and predator populations. Chassot et al. (2005) also stated that fishing generally affects species at higher trophic levels, which results in changes in their population dynamics and eventually alters the biomass of each component in the ecosystem.

CONCLUSION
Two Ecopath models of Bandon Bay were constructed, for 2007 and 2016. The main objective was to understand the changes in the bay's ecosystem after the inauguaration of the crab bank in 2007. Changes in most of the ecological indices revealed higher maturity and stability after 10 years of stocking by crab banks. Differences in abundance of each component between the two models were likely caused by fisheries. The bottom-up control of processes in the ecosystem of Bandon Bay was confirmed by the Ecopath model. Understanding the impacts of fishing activities on the ecosystem as well as examining likely top-down control processes (i.e., fishing control) in exploited ecosystem should receive focus for better resource and fisheries management of the productive Bandon Bay. Future work should also emphasise data quality and certainty of input parameters for better model performance.