The P. trituberculatus polyculture experiment was conducted using the land-based experimental enclosures in a pond. Enclosure is an efficient tool to investigate aquatic ecosystem, which establishes a relatively closed ecosystem by enclosing the water body, and there is no water exchange with the surrounding water (Sun et al., 2011). Land-based enclosure in this study means the enclosure built on the bottom of the pond to simulate pond ecosystem (Li et al., 1998a), which was different from the floating enclosure (Li et al., 1993). The experimental pond located in Ganyu County, Jiangsu Province, China (34°58′17.30″ N, 119°11′53.70″ E), covered 2 ha, with a water column depth of 1.6–1.7 m at the study site. Four land-based enclosures, representing four replicates of the same size (length × width × depth = 5 × 5 × 2 m), were constructed in the pond, which were lined with polyethylene (water-proof) and supported with wood poles. The structure of the enclosures was previously described in detail (Wang et al., 1998; Tian et al., 2001). At the bottom of the enclosure, the walls were covered with mud from the same pond, and supported by posts at 2.5-m intervals. An aeration system consisting of a blower, PVC tubes, gas tubes, and air stones was used for aeration and water circulation. Five air stones connected by a gas tube were suspended approximately 20 cm from the bottom in each enclosure. The land-based enclosure can satisfactorily simulate a pond and be suitably used for aquacultural and ecological experiments (Li et al., 1998). The entire experiment was carried out for 90 days from July 13 to October 13, 2014. Detailed descriptions of the enclosure are provided in the Supplementary Material(part 1. Location and layout of the experiment).

Juvenile swimming crabs (P. trituberculatus) were cultured with white shrimp L. vannamei and short-necked clam R. philippinarum in the four enclosures in the pond. All animals were purchased from Ganyu Jiaxin Aquatic Food Co., Ltd. (Ganyu, Jiangsu, China). The stocking densities of swimming crabs, white shrimps, and short-necked clams used in the four enclosures were 6 individuals/m², 45 individuals/m², and 30 individuals/m², respectively. Swimming crabs, white shrimps, and short-necked clams were stocked with an initial individual wet weight of 0.58 ± 0.10 g, 0.05 ± 0.01 g, and 0.91 ± 0.10 g, respectively.

Crabs cultured in the four enclosures were fed with chilled blue clam (Aloidis laevis) twice per day (06:00 and 18:00). The amounts of A. laevis supplied to the crabs were adjusted according to Zhou et al. (2010). Shrimps were fed a commercial pellet diet (Lianyungang Chia Tai Feed Co., Ltd., Lianyungang, Jiangsu, China), the amount of which was determined according to the product specifications. To assess growth and health condition of crabs and shrimps, 15–20 individuals per enclosure were sampled every ten days using a cage net (0.60 × 0.40 × 0.15 m) designed to trap crabs and shrimps in each enclosure; individuals were returned to the enclosure when sampling was completed.

The early, middle, and late periods of P. trituberculatus polyculturing were at 0–30, 31–60, and 61–90 d of the 90-d culture period, respectively. Three mass-balanced Ecopath models representing early, middle, and late culture periods of the P. trituberculatus polyculture ecosystem were constructed.

Ecopath models were constructed according to two master equations (Christensen and Pauly, 1992a; Christensen and Pauly, 1992b). The first equation represents the production of each functional group utilized in the ecosystem. Production is divided into predation, migration, biomass accumulation, exports, and other mortality, and is expressed as follows (Christensen et al., 2005):

where B_i represents prey biomass (i), B_j represents predator biomass (j), (P/B)_i represents the ratio of production to prey biomass (i), EE_i is ecotrophic efficiency, Y_i is catch biomass, (Q/B)_j is the ratio of food consumption to predator biomass (j), DC_ij is the ratio of biomass (i) being consumed by (j) to the total consumption of (j), BA_i is the prey biomass accumulation rate (i), and E_i is the difference between immigration and emigration of prey (i).

The second equation describes the consumption of a consumer group which is constituted by production, respiration, and unassimilated food:

where R_i is respiration and U_i is food which cannot be assimilated by consumers. The Ecopath model requires that at least three of the four parameters B, P/B, Q/B and EE are imported for each functional group. Because EE is difficult to obtain for each functional group, the EE value is typically estimated through the input of the other three parameters. Detailed descriptions of limitations, applicability, and methods of this model were published previously (Christensen and Walters, 2004).

We choose to simulate the dynamics of the polyculture ecosystem by constructing Ecopath steady-state models, rather than using dynamic simulation models such as an Ecosim model (Christensen et al., 2005). This is because aquaculture pond ecosystems are vulnerable to external disturbances such as weather changes. External disturbances may prevent Ecosims from accurate and comprehensive simulation of dynamic processes. With Ecopath, most of the model inputs can be obtained by field experiments directly. The modeled results may thus be more reliable than those of Ecosim models.

The basic input data and the estimated EE values of the three models are presented in Tables 3–5. The biomass of P. trituberculatus, L. vannamei, R. philippinarum, and benthic bacteria increased from the early to the middle and late period, while macro- and micro-benthos biomass showed a decreasing trend over these periods. The biomass of macro- and micro-zooplankton, bacterioplankton, and micro- and nano-phytoplankton increased from the early to the middle period, and then decreased from the middle to the late period. The biomass of pico-phytoplankton showed a marked increase from the middle to the late period. The groups of microbenthos and benthic bacteria had the lowest EE values in models of all three periods owing to the lack of consumers. The EE values of these two groups were all< 0.10. The periphyton also had low EE values in models of the early, middle, and late periods, with the values of 0.12, 0.11, and 0.04, respectively. Moreover, the EE value of pico-phytoplankton was also low in the model of the late period (0.24). The groups of P. trituberculatus, L. vannamei and R. philippinarum all had high EE values in the three models, and the EE values were higher than 0.90 in models of the middle and late period. The groups of blue clam A. laevis and shrimp feed also had high EE values (> 0.80, each) in models of all three periods. The detritus in sediment had high EE values in the early and middle periods (0.95 and 0.99, respectively). Other groups including microzooplankton and bacterioplankton had high EE values in the model of the late period (0.86 and 0.85, respectively).

Consumption by various groups in models of the three periods is presented in Figure 1. The total consumption increased from the early to the middle and late period. Benthic bacteria were the dominant consumers in all three culture periods, followed by L. vannamei. Bacterioplankton and macrozooplankton had the third and fourth largest consumption, respectively, in the early period, whereas bacterioplankton and P. trituberculatus had the third- and fourth-largest consumption, respectively, in both middle and late periods. Macrobenthos had the least consumption in all three culture periods, followed by microbenthos and microzooplankton.

Aggregating the three modeled ecosystems into discrete TLs led to the identification of five integer trophic levels in the early, middle, and late periods (Tables 6–8). TL I in the three modeled ecosystems was composed of primary producers, including micro-, nano-, and pico-phytoplankton; periphyton; and the detritus groups of A. laevis, shrimp feed, detritus in sediment, and detritus in water. 75.76%, 82.17%, and 76.77% of the energy on TL I originated from detritus groups in the early, middle and late periods, respectively, whereas the remaining 24.24%, 17.83%, and 23.23% originated from the primary producers in the early, middle and late periods, respectively. Energy flows at TL II encompassed almost all the consumer groups in the early, middle and late periods in this ecosystem, except microbenthos, which was more relevant at TL III in this ecosystem during all three periods. Most of the energy flow of the 17 groups in the early, middle, and late periods in this ecosystem was distributed at TLs I and II. Energy flow at TLs IV and V during the three culture periods was negligible. Benthic bacteria had the highest energy of flow to detritus in the early and middle periods (1493.45 and 1703.20 kJ m^-2, respectively), whereas in the late period, the highest energy flow to detritus was pico-phytoplankton (1950.37 kJ m^-2). The second-highest energy flow to detritus was pico-phytoplankton (532.80 kJ m^-2), macrozooplankton (638.61 kJ m^-2), and benthic bacteria (1914.31 kJ m^-2) in the early, middle, and late periods, respectively.

A Lindeman spine plot was constructed to represent flow and biomass for the models of the early, middle, and late culture periods (Figure 2). Primary production was calculated as 40.66, 60.38, and 54.84 kJ m^-2 for the models of the early, middle, and late periods, respectively, which contributed approximately 11.92%, 8.78%, and 11.13%, respectively to the TST. In the early, middles, and late periods, 891.60, 1001.00, and 1180.00 kJ m^-2 30 days^-1, respectively, of primary production were consumed by its consumers, whereas 1156.00, 1213.00 and 2582.00 kJ m^-2 30 days^-1, respectively, of primary production flowed to the detritus groups. In the early, middle, and late periods, 5,683.00, 8,857.00, and 10,843.00 kJ m^-2 30 days^-1, respectively, of total energy flow to detritus was consumed by primary consumers, whereas 150.55, 28.63, and 750.11 kJ m^-2 30 days^-1, respectively, of the total energy flow to detritus was accumulated in the group of detritus in sediment. The energy transfer efficiencies from TL II to III in the early, middle, and late periods was 4%, 4%, and 2%, respectively. The transfer efficiencies were lower than the 10% energy transfer efficiency of typical natural ecosystems (Lindeman, 1942). Most of the TST was distributed at TLs I and II in all three models.

The system attributes of the three models are described in Table 9. The TST in this ecosystem comprised total respiration, total consumption, and total energy flow to detritus; TST increased from the early to the middle and late period, with the values of 17,176.73, 25,211.79, and 33,803.86 kJ m^-2 30 days^-1, respectively. Total primary production, total production, and total biomass also showed an increasing trend from the early to the middle and late period. The ratio of total primary production to total respiration (TPP/TR) was 0.61, 0.43, and 0.50 in the early, middle, and late periods, respectively. On account of the high energy import via the addition of A. laevis and shrimp feed to support the growth of the cultured organisms, net system production had values of -1330.67, -2950.29, and -3752.76 kJ m^-2 30 days^-1 in the early, middle, and late periods, respectively. FCI was calculated as 20.38%, 17.90%, and 12.28% for the early, middle, and late periods, respectively. FML decreased from the early to the middle and late period, with values of 5.08, 4.89, and 4.50, respectively. Values of ascendancy showed an increasing trend from the early to the middle and late period. The A/C in the early, middle, and late periods was 41.51%, 42.41%, and 45.05%, respectively, and a pronounced increase was noted in late period. The O/C in the early, middle, and late periods was 58.49%, 57.59%, and 54.95%, respectively, and a marked decrease was noted in the late period.

During the 90-day culture period in the present study, the biomass of micro-, nano-, and pico-phytoplankton expressed varying fluctuating trends. The biomass of both micro- and nano-phytoplankton increased from the early to the middle period, and then decreased from the middle to the late period. In contrast, the biomass of pico-phytoplankton showed the opposite trend: it decreased from the early to the middle period, and then increased from the middle to the late period. The total phytoplankton biomass and primary production increased from early to middle and late periods. In line with Ban et al. (2015), inorganic nutrients (e.g., NH₄⁺, NO₃^-, and NO₂^-) increased remarkably from the early to the middle period in this P. trituberculatus polyculture ecosystem, considering the input of artificial foods. It was speculated that the increase in inorganic nutrients may have led to the increase in total phytoplankton biomass and primary production from the early to the middle period (Ramírez et al., 2005). The dominant phytoplankton community changed from nano-phytoplankton to pico-phytoplankton from the middle to the late period. This phenomenon may be correlated with the polyculture of R. philippinarum. The filter-feeding R. philippinarum efficiently transfers particulate matter present in the water column to the sediments by filtration coupled with biodeposition (Kautsky and Evans, 1987). Prins et al. (1998) and Souchu et al. (2001) reported that the consumption of phytoplankton by filter feeders could induce a shift in the phytoplankton composition towards phytoplankton with smaller cells and faster growth rates. R. philippinarum exhibits the highest phytoplankton consumption in the late period compared to the early and middle periods; thus, phytoplankton consumption by R. philippinarum may induce a change in the phytoplankton composition, mainly comprising pico-phytoplankton, which has a higher turnover rate. However, the low EE value (0.24) of pico-phytoplankton in the late period indicates that phytoplankton consumption by R. philippinarum could not suppress the growth of pico-phytoplankton, and that R. philippinarum was not food limited in this ecosystem.

The energy input in aquaculture pond ecosystems mainly comprises primary production and imported foods (Feng et al., 2015; Dong et al., 2017). Primary production and imported foods (blue clam A. laevis and shrimp feed) increased from early to late period in this ecosystem. Primary production accounted for 50.29%, 34.79%, and 39.71% of total energy input in the early, middle, and late periods, respectively. However, approximately 59.28%, 59.97%, and 73.29% of primary production was not utilized by consumers and flowed to detritus groups. Energy originating from detritus accounted for 83.00%, 87.00%, and 85.00% of total system throughput in early, middle, and late period, respectively. As a result, the polyculture ecosystem remained a detritus-dominated ecosystem throughout, which was in accordance with the results of Feng et al. (2017) and Feng et al. (2018a), showing that the detritus food chain is predominant in P. trituberculatus polyculture systems.

The group of benthic bacteria was the largest consumer over the entire culture period, thus the associated energy pathway of detritus in sediment to benthic bacteria was the main energy flow pathway. However, the energy flow structure showed some changing trends associated with organisms as total consumption by P. trituberculatus, L. vannamei, and R. philippinarum to total system consumption showed an increasing trend from early to middle and late periods, which accounted for 32.99%, 40.90%, and 48.64% of the total system consumption, respectively. Increased consumptions of these three functional groups led to increased energy flow to detritus during the entire culture period; moreover, pico-phytoplankton showed the highest energy flow to detritus in the late period. The total energy flow to detritus groups thus showed an increasing trend from the early to the middle and late period (4,221.29, 5,366.55, and 6,810.27 kJ m^-2 30 days^-1, respectively), and large amounts of energy were accumulated in the detritus group of detritus in sediment. Increased detritus accumulation was considered a potential risk to the ecosystem, which may cause deterioration of the bottom environment in a pond ecosystem (Lin et al., 2005). Management practices such as increasing bottom oxygen supply should be implemented to reduce this risk.

Biomass change of each functional group during the entire culture period could lead to changes in ecosystem structure, which in turn could lead to changes in ecosystem function. The indices of TST, AMI, A/C, and O/C are assigned to a number of groups that describe ecosystem properties in terms of ecosystem structure (Tomczak et al., 2013). In the polyculture ecosystem of the present study, TST, AMI, and A/C all increased from the early to the late period. The increase in TST indicated that the ecosystem size increased from the early to the late period. The increase in AMI indicated that the inherent organization, i.e. the degree of specialization of flow in the network of this polyculture ecosystem, increased from the early to the late period, which represents a trend of simplification of the food web structure (Latham and Scully, 2002). The A/C, which is the ratio of AMI to H, showed a marked increase from the middle to the late period, because of the decrease in H and increase in AMI, and this indicated that the ecosystem had the highest level of system organization in the late period. The system overhead represents the resilience of an ecosystem against external perturbations (Ulanowicz and Norden, 1990; Christensen, 1995). The O/C in the late period was lower than that in the early and middle periods, indicating that the ecosystem was less resilient in late period than in the early and middle periods.

FCI and FML are indices that describe the system attributes of energy flow (Tomczak et al., 2013). The values of FCI and FML decreased from the early to the middle and late periods. The fluctuating trends of FCI was contrary to that of A/C. which was in accordance with Baird et al. (1991), who demonstrated an inverse relationship between FCI value and A/C. The decreased FCI value indicated that the proportion of the TST that is recycled decreased from the early to the late period. Considering that the food composition of each consumer was almost unchanged during culturing, the decreased FCI may explain the decreased FML during the entire culture period. The decreased FML also indicated a simplification of the food web structure.

The polyculture system structure changed gradually to an ecosystem with larger size and higher level of organization from the early to middle and late periods. However, the overly increased TST and AMI may jeopardize the stability of ecosystem structure (Ulanowicz, 2003). It was demonstrated that the introduction of R. philippinarum may improve H through increasing total energy flow pathways (Feng et al., 2018a); improved H will decrease the AMI and increase system resilience (Ulanowicz, 2004). Moreover, as R. philippinarum was not food-limited in the late period, increasing the biomass of R. philippinarum may increase the EE value of pico-phytoplankton, thereby increasing the energy utilization of primary production. As a result, increasing the stocking density of R. philippinarum to maintain higher biomass may be an effective way to optimize such polyculture ecosystems.

Large amounts of detritus energy were accumulated in the group of detritus in sediment. Increased detritus accumulation may cause deterioration of the bottom environment of the pond ecosystem, which would further affect growth of organisms, especially the cultured organisms (Lin et al., 2005). In the present study, energy flow from pico-phytoplankton contributed the largest part of the total energy flow to the detritus. As the growth of phytoplankton was mainly supported by inorganic nutrients (Ban et al., 2015), introduction of macro-algae at a suitable biomass into this ecosystem may help absorb superfluous inorganic nutrients, thereby reducing phytoplankton production, detritus accumulation, and increase ecosystem stability. Moreover, macro-algae can also be used as aquatic products.