To assess the effects of wels catfish predation on birds, we utilised a control/impact approach, assuming that predation would be density-dependent (i.e. that it would be less significant in the control area, where predator density is much lower). We used a preliminary analysis of wels catfish stomach contents to guide our field sampling of their putative prey items. We then estimated the avian contribution to the diet of wels catfish in the Reserve using a stable isotope Bayesian mixing model and its biomass using electrofishing removal. Based on three different bioenergetic models, we estimated wels catfish daily feeding rates and used this information to estimate birds’ consumption by wels catfish in the Reserve (impact area) during the nesting and chick growing period (April-June, 90 days) of 2019. We then used mallard (Anas platyrhynchos) chick counts in 2017–2019 to compare reproductive success in the control and test areas, to gauge potential differences in predation magnitude and considered reproductive bird trends in the Reserve and surrounding areas to detect any broad effects.

Our control area consisted in the area where the Oglio River flows into Lake Iseo, about 17.5 km north of the Reserve. Both control and test areas share an equal number and type of other potential predators (e.g. birds or mammals), but the control area is characterised by high anthropogenic presence, no bird protection measures (i.e. hunting is allowed) and slightly deeper, flowing water. Wels catfish is present in the control area, but at much lower densities than in our test area (this study).

Our test area consisted in the Sebino peat bogs (‘Torbiere del Sebino’, in Italian), a marshland of ~ 360 ha, located near the southern shore of Lake Iseo (6530 ha), in northern Italy. These bogs are typically shallow (average depth 5 m, this study) and cold-temperate (5.8–28.3 °C during 2019, this study), with abundant emergent and submerged aquatic vegetation and are intermittently connected to the Lake. The Lamette part of the bogs is a shallow (max. depth 5 m) marshland with abundant reeds that has the closest connection with the Lake and is a strict Reserve (i.e. no human activities are allowed). Conversely, the ‘Torbiere’ and ‘Lama’ parts of the bogs are a series of deeper (max. depth 9 m) lakes, have restricted public access and, in some parts, recreational fishing is permitted (Fig. 1). The Sebino Peat Bog Nature Reserve (hereafter referred to simply as the Reserve) was established in 1984 and is protected under both national and European legislation (Natura 2000 network). Prior to this protection, the bogs were used extensively for peat harvest. The Reserve is a key nesting area for waterbirds, as well as an important stopover area during the migration period and a relevant wintering site (Trotti 2019). Out of the total 268 bird species observed in the Reserve since the 1950s, 14 are introduced and 31 are covered by the EU Birds Directive (Trotti 2019).

Figure 1.

Location of the study area in Italy (upper left panel, in red) and location of the control (in green) and test (in orange) areas at the opposite ends of Lake Iseo (upper right panel). The lower left panel depicts the control area (in green), where the Oglio River enters Lake Iseo. The lower right panel depicts the test area, the Sebino Peat Bogs (Torbiere del Sebino), a Nature Reserve declared in 1984. Three different areas of the Reserve are named and shaded in yellow, blue and magenta, based on their depth, vegetation and numbers of human activities permitted. Satellite and aerial imagery are from Google Earth.

Wels catfish were first accidentally introduced to Lake Iseo through the Oglio River and ultimately to the Reserve in the late 1980s (Mancini, unpublished data), but their numbers in the Reserve were initially low and they were not detected until much later. Wels catfish biomass in the Reserve is likely to have increased around 2005–2008, leading first to their detection and then to selective harvest in 2011, to limit the introduced fish population in the Reserve (Mancini, unpublished data).

We assumed that small-sized catfish would not be effective predators of adult and sub-adult birds, due to mouth gape limitations. Therefore, we sampled 31 large-sized wels catfish (total length > 98 cm, hereafter simply referred to as length) by spearfishing in the Reserve during spring-summer 2019. These individuals were analysed for stomach contents using a volumetric point method (Windell 1968) to collect preliminary information on catfish diet and guide sampling of putative prey for stable isotope analysis. Not unexpectedly for predator fish, 20 wels catfish stomachs were found empty and 11 stomachs contained prey, of which four contained red swamp crayfish (39.1% of total combined volumetric contents), three contained unidentified remains (13%), two contained fish of different sizes (Italian rudd, Scardinius hesperidicus and unidentified, 13%), one contained a rodent (unidentified, 8.7%), one a feather and one an adult bird (cormorant, Phalacrocorax spp., 26.1%).

Based on this guidance and on literature dietary information (e.g. Copp et al. 2009), we sampled catfish muscle from our preliminary sample and a range of putative prey species from the Reserve for stable isotope dietary analysis. Permits for collection of field samples were obtained from the ‘Torbiere del Sebino’ Nature Reserve administration within the wels catfish control programme, all individuals were immediately euthanised after capture and no protected species were culled in this study. We collected muscle of putative aquatic prey, including large piscivorous fish (three individual samples from two species), small generalist fish (12 individual samples from six species) and alien red swamp crayfish (Procambarus clarkii, eight individual samples) through electrofishing. We also collected muscle from putative terrestrial prey, including one rodent muscle (Rattus sp. from fish stomach contents) and feather (calamus) samples from several waterbird species (43 individual samples from eight species, from fish stomach contents, birds found dead and natural moulting) through an opportunistic collection. A detailed list of samples and species is provided in Suppl. material 1: Table S1. Collecting feathers allowed for non-lethal sampling of birds, while offering a stable isotope measure comparable to muscle tissue (Hobson and Clark 1992a, b).

Tissue samples were dried at 60 °C, ground to fine powder (muscle) or cut to size (feathers). As lipid variations in tissues can alter d¹³C values, feathers were rinsed in 2:1 chloroform/methanol solution to remove surface lipids and stable isotopes ratios of C in other tissues were later arithmetically corrected for lipid content (Kiljunen et al. 2006). Samples were analysed for stable isotope ratios of C and N at the University of Jyväskylä, Finland, using a Thermo Finnigan DELTA^plus Advantage continuous-flow stable isotope-ratio mass spectrometer (CF-SIRMS), coupled with a FlashEA 1112 elemental analyser (Thermo Electron Corporation, Waltham, MA, USA). The resulting ratios were expressed in terms of relative concentrations relative to a laboratory reference standard. While a rodent and a feather sample collected from stomach contents were isotopically analysed, they were eventually excluded from further analysis to avoid issues with potential consumer contamination and low sample size.

We estimated diet proportions of wels catfish with a Bayesian mixing model under R statistical software 3.6 (R Core Team 2019) through the mixSIAR package (Stock et al. 2018), which accounts for sample size when estimating diet proportions and using literature isotope fractionation [δ¹³C = 0.4 ± 1.2, δ¹⁵N = 2.3 ± 1.6, (McCutchan et al. 2003)]. We tested for wels catfish size effects by running a mixing model where catfish length was considered as a continuous covariate, after testing that differences in diet composition between size classes suggested by our preliminary stomach content analysis were not significant in the isotopic space (PERMANOVA P-value > 0.05). We also used our preliminary stomach content analysis and expert judgement to produce informative priors (see Suppl. material 1: Fig. S3) that were used in the mixing model (Moore and Semmens 2008), with a very conservative approach to bird consumption values. We additionally ran re-sample simulations (100 iterations each for samples sizes 2–50) to assess isotopic sources sample size effects on the results of the model, using the package ‘samplesim’ (Casajus et al. 2021). Putative prey species were divided into five functional groups (Small fish, Large Fish, Crayfish, Invertivorous Birds, Herbivorous Birds and Piscivorous Birds), based on taxonomy and ecology. The piscivorous birds included cormorants and grebes (thus with both a marine and freshwater signal), the herbivorous birds included swans, mallards, pochards and moorhens and the invertivorous birds included water rails. Chicks of most waterbird species are invertivorous during growth (Sugden 1973), but no specific groups were created for chicks.

To estimate wels catfish biomass in the Reserve, we used 73 boat-mounted electrofishing events between 2013 and 2019, covering the shoreline of all three main areas of the Reserve (‘Lama’, 40 events, ‘Lamette’, 23 events, ‘Torbiere’, 10 events) and all seasons (but focusing on autumn and spring). A total of 1356 individuals were caught and removed from the Reserve using electrofishing, for an overall weight of 8113 kg. Wels catfish were of average total length 84.45 cm (median 82 cm, min. 2.2/max. 211 cm) and of average weight 5.98 kg (median 3.84 kg, min. 0.015/max. 92.75 kg).

We estimated the overall biomass of wels catfish in each of three areas of the Reserve, by averaging the detected density at each sampling event. Density was calculated as a function of biomass harvested and area sampled during each event, where area was the length of the shoreline fished, multiplied by the effective radius of the electrofisher (i.e. 5 m). Density trends over time were analysed with linear regressions. Given that we only estimated density in shoreline areas and that electrofishing catchability is high but not perfect, ours was likely an underestimation of total biomass.

To estimate wels catfish biomass in the control area, we used a boat-mounted electrofishing survey followed by three visual census surveys, carried out between 2012 and 2016 and spanning from April to July, along the shorelines of the control area (including the terminal part of the Oglio River).

We estimated wels catfish annual consumption of prey by developing a specific bioenergetic model for the local conditions and sampled size cohorts. We used the Wisconsin bioenergetic model (Hanson et al. 1997), parametrised according to experimental studies on wels catfish and similar species (Hilge 1985; Raat 1990; Xiao‐Jun and Ruyung 1992; Bourret et al. 2008, see Suppl. material 1: Table S2) to estimate annual consumption. Fish being poikilotherms, the model relies on water temperature to assess metabolic rates and on prey energy content to model body mass gain. We thus used records of daily average temperatures in the Reserve in 2019 (when catfish were sampled, see Suppl. material 1: Fig. S1), food energy contents from Cummins and Wuycheck (1971) (see Suppl. material 1: Table S3) and diet composition derived from our stable isotope analysis as inputs to the model. We used model results to calculate an average daily consumption of wels catfish during the chick growing period, accounting for site-specific diet composition, size and growth (see Suppl. material 1: Table S3), estimated based on yearly length increments from Rossi et al. (1991). Weight increments were derived from length, using a weight/length regression fitted to our data (W = 7E-05*L^2.6535, R² = 0.9374).

Additionally, we compared our model results with two previous consumption estimates. An average daily consumption of 1.99% of wet mass day⁻¹ was estimated by Omarov and Popova (1985) and Orlova and Popova (1986) for wels catfish in Georgia over the whole year. A similar value of 1.32% of wet mass day⁻¹ was also estimated by Copp et al. (2009) in their review, also including European data.

We then used the estimated biomass and the three different estimates of average daily consumptions listed above to calculate the total quantity (kg) of prey ingested over a year. The consumption of bird prey was then estimated, based on its dietary proportion, as estimated by our stable isotope mixing models, accounting for the error in the dietary contribution (50% confidence interval) and in the biomass estimate (SD around the mean) when calculating the upper and lower confidence limits of the consumption estimate. Bird consumption was estimated exclusively for the > 98 cm size cohort, as diet was determined only for this size class.

We used nesting bird surveys from 2002–2019 (Brichetti 2002; Longo 2009; Trotti et al. 2016; Trotti 2017, 2018, 2019) to assess the consistency and short-term trends of waterbirds (i.e. ecologically dependent upon wetlands) nesting in the Reserve, which could be potential prey for wels catfish. Surveys were carried out by 1–3 experienced observers, between May and September, for a total of more than 30 days per year. Surveys included nestling and chick counts during and after the breeding season and were performed from both shore and boats, at dawn and dusk. We focused on nesting birds, as young birds are smaller and likely to be more readily predated and the breeding season coincides with the period of warmer temperatures, which increase fish activity and thus predation. Given the few data points, trends in nesting birds numbers were identified using simple linear regressions and given a qualitative classification as Stable (coefficient <\0.2\), Moderate increase/decrease (\0.2\ < coefficient <\0.5\) and Increase/Decrease (\0.5\ < coefficient), reporting only R² values. Additionally, long-term trends in wintering numbers estimated by the International Wetland Census for the functional/ecological spatial wintering unit that includes the study area, plus contiguous spatial units (Zenatello et al. 2014) were also retrieved for comparison of local and larger spatial scale trends.

To offer a comparison of the scale and potential impact of wels catfish consumption on birds, we used average weights of each bird species (accounting for sexual dimorphism in size, see Suppl. material 1: Table S4) to estimate an overall nesting adult biomass in the Reserve during the summer season (i.e. the season when fish predation is most likely). We may have underestimated the total waterbird biomass because we did not include the weight of chicks and there may be constant turnover of individuals in the bird population. Furthermore, the waterbird biomass would be considerably higher in winter, when bird numbers increase.

Finally, we used reproductive success surveys, carried out in 2017–2019 with similar methods as the nesting bird surveys, to assess differences in the number of chicks per couple of waterbirds in the control and the Reserve areas. We chose mallard as a test species, as it is a cosmopolitan and abundant species in both areas and counted the number of chicks per couple in early (mid-April/mid-May) and late (June) stages of the chick growing period in order to test differences in the median number of chicks at the late stage across the areas and differences in the slopes between early and late stages across areas, under the null hypotheses that different areas would have equal means and slopes. We used the non-parametric Mann-Whitney test on the medians and the test on the difference between the slopes from two independent samples outlined in Howell (2012).

Wels catfish in our sample (n = 30, mean 139.9 cm, SD 30.5 cm, length range 98–191.5 cm) were generally spread between sources in isotopic space (Fig. 2), but there were no clear trends stemming from wels catfish length in the mixing model.

Figure 2.

Isotopic space positions of wels catfish and its putative prey sources in the Reserve, corrected for isotopic fractionation. Error bars represent standard deviations of each prey source. Feathers were analysed for birds and muscle for other taxa; both were corrected for lipid content.

The Bayesian mixing model indicated that birds composed 12.2% (5–27.9%, 50% confidence interval) of the diet of wels catfish > 98 cm in length. More specifically, invertivorous bird prey composed 8.6% (4.6–14.1%, 50% confidence interval) of the diet, while herbivorous and piscivorous bird prey composed 1.7% (0.2–6.4%, 50% confidence interval) and 1.9% (0.2–7.4%, 50% confidence interval) of the diet, respectively (Fig. 3). According to the mixing model, larger catfish did not have a higher proportion of birds in their diet.

Figure 3.

Curves of dietary proportions of wels catfish prey sources, derived from the Bayesian mixing model for stable isotopes including weakly informative dietary priors.

Our sample size simulation indicated that increasing sample size up to 50 samples was likely to downplay the dietary proportions of crayfish and large fish and increase the relevance of birds (particularly piscivorous birds) by up to ~ 5%, but not to decrease the width of confidence intervals, except for invertivorous birds (see Suppl. material 1: Fig. S4a, b). Doubling wels catfish sample size would have similarly yielded a ~ 10% increase in the relevance of birds (particularly invertivorous birds) and a similar decrease of the width of confidence intervals (see Suppl. material 1: Fig. S4c).

Detected wels catfish density ranged 3.4–174 kg/ha, with highest densities recorded in the ‘Lama’ part of the Reserve and no clear trends were observed in 2013–2019 (see Suppl. material 1: Fig. S2). The estimated average total biomass of wels catfish (of all sizes) along the Reserve shorelines was 1237 (± 254) kg. Despite being only 22.9% of the total individuals sampled, fish 98–191 cm of length were estimated to have an average biomass of 1024 kg, 83% of the total biomass.

None of the surveys in the control area was able to detect wels catfish, despite covering a total combined surface of 75.7 km². We thus conservatively concluded that, albeit present in Lake Iseo, wels catfish density in the control area was below detectable levels and, thus, likely to be negligible compared to the biomass detected in the Reserve.

Our bioenergetic model suggested an average daily ratio of 1.7% wet mass day⁻¹ for a wels catfish > 98 cm during the chick growing period (and an average daily ratio of 1.5% wet mass day⁻¹ over the whole year).

The estimated average bird consumption for the wels catfish population > 98 cm during the chick growing period was 224, 148 and 187 kg, respectively, as estimated through the three different daily ratios (Fig. 4). By comparison, the overall biomass of the 243 waterbird nesting pairs in the Reserve during 2019 was estimated at 792 kg (691 kg excluding adult swans, Fig. 4).

Figure 4.

Estimated consumption of birds by wels catfish in the shoreline of the Nature Reserve during the chick growing period of 2019, as obtained with the three different estimates of daily consumption ratios. Error bars account for uncertainty in both dietary and biomass estimations. The horizontal dashed lines indicate the estimated biomass of nesting adult waterbirds observed in the Reserve during the chick growing period of 2019.

A total of 12 waterbird species that could be potential prey of wels catfish were found nesting in the Reserve in 2002–2019, for a total of 243 nesting pairs. The most abundant breeding species was the great cormorant (Phalacrocorax carbo, 83 breeding pairs), while the least abundant was the mute swan (Cygnus olor, five breeding pairs). The number of breeding pairs was generally consistent in 2017–2019 for most species (Table 1), increasing for some species (e.g. the great cormorant), but markedly lower than those detected in 2002 for other species (e.g. Eurasian coot, Fulica atra). The total estimated nesting waterbird biomass in 2019 was 792 kg (691 kg, excluding adult swans which are particularly large prey).

During 2017–2019, mallard reproductive success at early stages of the chick growing period was equal in the test rather than in the control area (Fig. 5, Mann-Whitney P-value > 0.7 for all years). However, in 2018 and 2019, there were statistically significant differences between the test and control areas in the number of chicks at late stages of the growing period (Mann-Whitney P-values < 0.01 and < 0.05, respectively), even if those were not significant in 2017 (Mann-Whitney P-value = 0.06). The differences in slopes expressing the rate of decline were statistically significant in 2018 (P-value < 0.03), but not in 2017 and 2019 (P values 0.3 and 0.06, respectively).

Figure 5.

Mallard chicks per pair in the control (green) and test (orange) areas, as detected in early (mid-April/mid-May) and late (June) stages of the chicks growing period. In boxplots, black horizontal lines represent medians, boxes represent the first and third quartile, whiskers represent minimum and maximum values and dots represent extreme values.