Project 1 (Dec 2014–May 2016): Biodiversity and biological interactions in six Singapore reservoirs — a pilot study of food web and trophic structure and implications for environmental and water quality management.

Project 2 (June 2016–Oct 2018): Biodiversity and biological interactions in Singapore’s reservoirs and waterways — a study of food web and trophic structure and implications for environmental and water quality management.

Darren C J Yeo, Heok Hui Tan, Timothy Jardine, Jeffrey T B Kwik, Rudolf Meier, Jia Huan Liew, Clare Wilkinson, Rayson B H Lim, Claudia L Y Tan, Ming Li Chen, Wen Qing Ng, Yvonne Y W Kwang, Abel C Y Saw and Shan Shan Liu

The data presented in this paper were collected from 12 man-made reservoirs in the tropical island of Singapore (1°21.0' N, 103°49.11' E; Fig. 1, Table 1), six being sampled in Project 1 and six sampled in Project 2. The reservoirs are located within catchments with land-use regimes ranging from protected nature reserves (i.e. riparian vegetation consisting of secondary and primary forests) to urban centres (i.e. riparian zone consisting of rocky rip-rap and impervious surfaces with minimal herbaceous plant cover). The reservoirs vary in age (i.e. time since impoundment), ranging from eight to more than 100 years (Yeo and Lim 2011). The reservoirs are typically characterised by warm (mean water surface temperature of 28.9–31.0°C), slightly turbid and basic (mean pH from 7.1–8.4) waters that are dominated by non-native fish assemblages (Kwik et al. 2020, Tan et al. 2020).

Figure 1.  

Location of the 12 reservoirs sampled in Singapore.

This project was funded by PUB, Singapore’s Water Agency [National University of Singapore grant number R-154-000-619-490 and R-154-000-A20-490].

Each of the 12 reservoirs were surveyed independently for 2–3 months between December 2014 and October 2018 in Singapore. Our surveys targeted macrofauna (i.e. fish, benthic invertebrates), microfauna (i.e. zooplankton, phytoplankton and periphyton) and riparian vegetation (i.e. C3 and C4 plants) that were broadly representative of the biotic communities observed in each reservoir (Table 2). In total, the 12 food webs sampled constituted 52 fish species and one hybrid (Oreochromis mossambicus × Oreochromis niloticus), 15 isotopically distinct invertebrate orders, classes or families and nine basal resources.

Fishes were collected using a combination of cast netting (net dimensions: 4 m radius, 2 cm mesh size), trapping (trap dimensions: 50 cm × 60 cm × 40 cm, 2 cm mesh size) and boat electrofishing (pulsed DC electrofishing, model ETS-MBS-1D-COL) in the littoral zone of each reservoir to optimise sampling coverage across various depths and fish sizes. We performed 10 casts, deployed three traps (for 48 hours) and conducted four 5-minute bursts of electrofishing, per transect. Live specimens were identified to species level (Baker and Lim 2008, Ng and Tan 2010), measured for standard length (cm), weighed for total wet weight (g) using PESOLA weighing scales and up to three individuals per species were euthanised for GCA and SIA. Invertebrate samples were collected using multiple methods. Specifically, benthic invertebrates were collected using two submerged colonisers filled with exogenous substrate (i.e. coconut palm fronds) which were deployed at a fixed depth of 1–1.5 m in the littoral zone for a total of 8–12 weeks at each transect (Loke et al. 2010). Additionally, a hand-held dip net (500 µm mesh size) was used to collect invertebrates from the surface and plankton nets (100 µm mesh size) used to collect zooplankton. All fixed (in 70% ethanol) invertebrate specimens were identified to the highest taxonomic resolution possible, based on identification keys to the freshwater macroinvertebrates of the Malaysian region and of Singapore (Yule and Yong 2004, Blakely et al. 2010). Phytoplankton were also collected by towing a 80 µm and 50 µm mesh size plankton net at the surface for ≈ 100 m per transect. Zooplankton and phytoplankton samples were then separated manually using pipettes or by successive filtration using cell strainers (100 µm, 70 µm and 45 µm mesh size) and the purity and identity of the samples evaluated under a dissecting microscope. Other basal resources were collected: periphyton and epiphytic algal samples were sampled from hard surfaces in the littoral zones; and macrophytes and riparian plants and/or grass samples comprised leaf clippings. 

To facilitate representative sampling, each reservoir was systematically subdivided into six or eight transects (dependent on reservoir size; ≈ 200 m per transect) spanning multiple littoral habitat types (i.e. rocky bund, forested, macrophyte-dominated). Sampling effort and protocols for the various taxonomic groups were standardised to facilitate comparability between reservoirs as described above. In addition, fish species identity was validated by taxon experts from the Lee Kong Chian Natural History Museum.

We followed general protocols from Liew et al. (2018) to combine GCA and SIA data to construct empirical food webs for all 12 reservoirs. We used GCA to provide information for the selection of potential prey, before estimating the relative strength of trophic relationships using Bayesian Mixing Models. 

I. Gut content analysis

The gut content of 47 fish species with at least four individuals having full gut were analysed to complement the stable isotope analysis (Kwik et al. 2020). In total, 1269 suitable guts were included for the dietary analysis. Dietary items for each fish species were identified under a dissecting microscope in a Petri dish and grouped into ten broad categories: 1) substrate (i.e. inorganic sediment), 2) unidentifiable animal matter (i.e. highly-digested matter of non-fish vertebrates), 3) plant matter (e.g. whole or parts of leaves, fruit and flowers), 4) periphyton (i.e. benthic and filamentous algae mats), 5) phytoplankton (i.e. pelagic algae, e.g. Microcystis, Planktolyngbya and Staurastrum), 6) zooplankton (e.g. Rotifera, Copepoda), 7) insect larvae (i.e. whole or parts of aquatic insect larvae, e.g. Chironomidae, Caenidae), 8) gastropods (i.e. aquatic snails and bivalves, e.g. Viviparidae, Ampullariidae), 9) decapod crustaceans (i.e. crabs and shrimp, e.g. Palaemonidae) and 10) fishes (i.e. whole body, scales or bones). The relative contributions of various dietary items for each fish species were determined, based on their frequency of occurrence (FO_%) and volumetric contribution (VO_%) to calculate the feeding index (Sá-Oliveira et al. 2014):

\(FI_i = \frac{(FO_i \times VO_i)}{\sum (FO_i \times VO_i)}\)

where FI_i = feeding index of species i, FO_i = frequency of occurrence of diet item i and VO_i = volume of diet item i.

II. Stable isotope analysis

We estimated the relative strength of trophic interactions of consumers and resources in the food webs using bulk ¹³C/¹²C (i.e. δ¹³C) and ¹⁵N/¹⁴N (i.e. δ¹⁵N) isotope profiles. Tissue samples for primary producers (e.g. riparian plants and macrophytes, phytoplankton, periphyton) consisted of leaf clippings from plants, filtered phytoplankton samples and substrate-free periphyton samples. Invertebrate samples comprised whole organisms for smaller specimens (e.g. dipterans) and muscle tissue for the larger taxa (e.g. gastropods). For fish samples, we extracted muscle tissue from the dorsal region of each individual. We collected a minimum of three samples per taxon and excluded rare species (with less than three individuals collected) from subsequent analyses. However, we made an exception for taxa that were abundant in the study site, but were difficult to isolate for SIA (e.g. Copepoda).

All samples were oven-dried for 48–72 h at 68.5–70.0°C, homogenised, ground to a fine powder and weighed (to the nearest 0.0001 g): 1 mg for consumers (e.g. fish, invertebrates) or 4–5 mg for primary producers (e.g. phytoplankton and plants) following protocols described in Jardine et al. (2003). In total, 5339 tissue samples were prepared and analysed. These consisted of 1042 primary producers, 1476 invertebrates and 2821 fish samples, respectively. These were packed in standard tin capsules and sent to the Stable Isotope Facility at the University of California, Davis for determination of stable isotopic profiles (δ¹³C and δ¹⁵N). 

III. Construction of empirical food webs 

We summarised the trophic information derived from GCA and SIA into predation matrices (n = 12). The GCA information was used to inform and/or restrict the pool of potential prey included in the models for each fish species, while information from published literature were used to identify potential resources for the invertebrate taxa. We used Bayesian stable isotope mixing models to estimate the proportional source contribution to diets of consumers by fitting probability models to the isotopic data (e.g. isotopic ratios, elemental concentrations, sample variations and trophic fractionation). 

Before running the mixing models, we corrected the δ¹³C isotopic profiles of samples that comprised whole individuals (e.g. small invertebrates) for lipid-enrichment using procedures described in Logan et al. (2008). We also adjusted the stable nitrogen and carbon isotope data to account for trophic discrimination, using taxon-specific values for δ¹⁵N (following Bunn et al. 2013) and generic values for δ¹³C (0.4 ± 1.3‰; Post 2002). We note that fractionation values reported in Bunn et al. (2013) were derived from lotic ecosystems, while our study systems are lentic. Nevertheless, the range of fractionation values fall within commonly reported fractionation ranges of 2–4‰, so we did not expect our findings to deviate significantly if we used lentic-specific (but not taxon-specific) values instead. Moreover, strong evidence of inter-taxa differences in δ¹⁵N fractionation (Sweeting et al. 2007, Bunn et al. 2013) suggest the importance of accounting for taxon-specific variations in isotopic fractionation, especially given the broad taxonomic coverage of our work. We ran 30,000 iterations of each mixing model (i.e. each consumer) with a burn-in of 10,000 draws using Markov Chain Monte Carlo (MCMC) processes on the simmr statistical package, version 0.3 (Parnell 2020). Model convergence was assessed using Gelman-Rubin diagnostic parameter (≈ 1).

We assessed the feasibility of all our mixing models prior to extracting finalised source contribution values using two criteria. First, we ensured that consumer isotopic profiles were bounded within mixing polygons (evaluated by an isoplot produced by the model; Phillips et al. 2014). The mixing polygons were also used to assess isotopic overlap between taxa in a food web. If there was significant isotopic overlap (ANOVA/t-test; p-value < 0.05) between phenotypically similar taxa, taxa were pooled to a higher taxonomic level, for example, genus for fish and to the order or class level for invertebrates, to create isotopically distinct taxonomic units. Second, we ensured that sources (i.e. prey) warranted inclusion in mixing models by removing items which fall below a minimum threshold. We used the following equation to calculate consumer specific threshold values: (1/N)*0.25, where N is number of sources for a consumer. If multiple food sources fall below the threshold value, we deemed them as ‘unimportant’ sources and removed them sequentially (starting from the lowest contributor), while ensuring that the removal of each source does not result in consumer profiles falling outside mixing polygons. We also checked for missing consumer sources by assessing respective mixing polygons, finding that a primary source was missing in three reservoirs where phytoplankton were too scarce to meet the required weight for SIA. For these three reservoirs, we created an ‘unknown producer’ source node using the common practice of adapting isotopic signatures of zooplankton, a likely predator of this missing resource (Grey et al. 2000, Post 2002, Matthews and Mazumder 2003). Finally, we summarised source-contribution for each consumer using the median value of posterior distributions into a predation matrix (column = predator; row = prey). 

We provide the R script used for producing the empirical food webs, as well as the raw data from one reservoir (res 6) to facilitate replication of our procedure (Suppl. materials 1, 2) and produce a food web (Fig. 2). 

Figure 2.  

Food web diagram for Reservoir 6. Created from stable isotope data, using the code in Suppl. material 1 and data in Suppl. material 2. The size of each node representing individual taxa is indicative of the number of links formed with other taxa (larger nodes indicate more links), while the thickness of arrows representing predator-prey relationship indicates the relative interaction strengths (thicker arrows indicate greater interaction).

N.A. (already included in “Study area description”)

Table 2