Estimation of Fishery Losses from Great Cormorants during the Wintering Period in Greek Lagoons (Ionian Sea, W. Greece)

Abstract

The present study aims through a modeling approach to quantify fishing losses from the impact of great cormorants (Phalacrocorax carbo sinensis) during their wintering period in Greek lagoons. A number of assumptions were incorporated into the model regarding fish population growth, species distribution, age (or size) of fish caught, and the different fishing strategies that could be applied in the studied lagoons. The results indicated that the mean value of daily economic losses ranged from 0.614 to 1.075 €·bird⁻¹·day⁻¹, whereas the ratios of biomass losses to landings biomass and of economic losses to economic profit ranged from 0.18:1 to 3.80:1 and from 0.14:1 to 4.18:1, respectively, depending on the lagoon. The results supported a strong competitive relationship between great cormorants and fisheries in lagoons of the Amvrakikos Gulf.

1. Introduction

Coastal lagoons are crucial components of local people’s cultural heritage, coastal environment, and economic vitality [1]. They serve as transitional areas between the open ocean and inland waters and are characterized by environmental fluctuations (for a review, see [2]). Given that these systems are habitats with an increasing concentration of fishing activities, because several fish species use them as part of their life histories for spawning, feeding, and refuge [3], the majority of them are protected by international conventions for the conservation of biodiversity (e.g., Natura 2000). Around the Mediterranean, an area of at least 6500 km² of coastal lagoons [4] is exploited as fishing grounds by local fisher associations [5]. The Greek lagoons cover a total surface of about 350 km² [6] and fishery exploitation is a common extensive culture, based on seasonal ongoing migration movements of fry and adult euryhaline fish species between the sea and lagoons (e.g., mugilids: flathead grey mullets Mugil cephalus Linnaeus, 1758, thicklip grey mullet Chelon labrosus (Risso, 1827), leaping mullet Chelon saliens (Risso, 1810), golden grey mullet Chelon auratus (Risso, 1810), and thinlip grey mullet Chelon ramada (Risso, 1827); gilthead seabream Sparus aurata Linnaeus, 1758; European seabass Dicentrarchus labrax (Linnaeus, 1758); and European eel Anguilla anguilla (Linnaeus, 1758)). Fishery production mainly comprises mugilids (about 56%) [6].

Aquatic birds can be seen as an integral part of these transitional aquatic systems, where an ongoing conflict with human activities exists. In fact, bycatch, a global issue, is one of the most significant factors contributing to the decline of aquatic bird populations (e.g., [7,8,9]). It is estimated that 400,000 birds per year are incidentally caught in gillnets globally [10], and in some cases, this taxon shares common resources with the fishers [11,12,13,14,15,16,17,18]. In contrast, in the lagoons of the southern Baltic Sea in Poland, the dominant piscivorous bird species, the Goosander, does not negatively impact fishing activities because it primarily preys on small fish species that are not the target of fishers [19].

The great cormorant Phalacrocorax carbo sinensis (Linnaeus, 1758), hereafter cormorant, is a species of piscivorous aquatic bird whose population has considerably increased throughout Europe in recent decades. This increase is due to protection (EU Directive 79/409, Directive 2009/147) and to increased food availability leading to eutrophication of aquatic habitats and commercial fish stocking [11,13,20,21,22]. The European population is estimated at 401,000–512,000 pairs, which equates to 803,000–1,020,000 mature individuals [23]. In the southern wetlands of Europe, the cormorant appears year-round but in greater numbers during the winter period [24,25,26,27]. In Greece, the species’ breeding population has increased over time [27], which is accompanied by an increase in its population on wintering grounds [28]. There are six cormorant colonies in Greece, with a total breeding population of 5600 pairs, while in winter, their population amounts to approximately 22,000 individuals [29].

Studies on the cormorant’s diet, energy requirements, impact on fish populations and fisheries, ethology [13,30,31,32,33], and management issues [13,28,34,35] have been extensively carried out worldwide. On the other hand, the fish-eating habits of the species and its increased numbers have caused severe conflicts with fisheries activities in many countries [21,30]. Cormorants’ diet is described as opportunistic, because they do not select specific species or sizes of fish [24] but rather focus on fish species available in the ecosystem in which they settle. The European cormorant population is estimated to consume about 1000 t of fish per day [11] and severe economic losses are recorded in freshwater fish farms ranging from 150 to 1500 €·10⁻¹ ha⁻¹ [11,12,17]. In the Czech Republic, the annual losses of carp pond aquaculture for 2019 are estimated to be EUR 4 million [36], while in a few countries/regions (e.g., Belgium (Waloon), Finland, Romania, the Czech Republic, Slovakia, Germany (Saxony), and Latvia), recognizing the conflict between cormorants and aquaculture has taken the form of financial compensation or subsidies programs [12]. In the northern Baltic Sea, there is an ongoing debate about the role of cormorants in the coastal ecosystem, where some studies suggest that cormorants can regulate fish populations [13,14,16] or that the effects have been considered to be site dependent [15] or have no effect [37].

In Greece, the effects of cormorants on fisheries range from minor (in lakes) [25,38] to relatively significant (in lagoons) [24,26]. Cormorants often prey on fish species with low commercial value, but this pattern appears to be reversed when colonies of fish-eating birds are concentrated in habitats with high fish concentrations, such as lagoons [26]. In Greek lagoons, Mugilidae comprise 60–70% of daily preyed biomass from the species [24,26], revealing a possible strong conflict between fisheries. However, the above assessments must be considered under-valued, because they estimate the direct losses by birds which are composed of fish individuals with sizes smaller than the commercial size [13,16]. If cormorants prey upon substantial number of fish at younger age classes, the yields of the fishers might be affected after a time delay [39]. This is anticipated to be more noticeable in a large culture, such as lagoons [11,12,17], than in coastal fisheries [13,14,15,16,37], justifying the fisher’s perception of the negative impact of cormorants in lagoon fishery production [24].

In the present study, we estimate the catch losses due to cormorants’ predation on fish resources during their wintering period in the lagoons of the Amvrakikos Gulf, using modeling techniques that estimate the loss in biomass and income based on individuals preyed upon by the cormorant population when the fish enters the exploited phase. The model used is a common population dynamics model [40] incorporating an economic component, which has been estimated for grey mullets. The basic idea of the model is that the fish size of grey mullets preyed on by cormorants in lagoons of western Greece [24,26] is smaller than the minimum length size of capture [41], and therefore, the effect of fish prey will be obvious as losses in fishery production after a time lag [39]. Moreover, in order to cover most of the uncertainties of input variables, the model was run for 60,000 iterations where the input variables, randomly, varied between a range (from available information), delineating the cormorant effect on lagoon fisheries as an expected range of values (biomass or income).

2. Materials and Methods

2.1. Site and Data Scources

The Amvrakikos Gulf (Western Greece, Ionian Sea) is a fjord-like hydrological regime (400 km²) [42] around which 15 lagoons are located (Figure 1), covering a total area of about 96.2 km² that are protected under the Ramsar convention (www.ramsar.org (accessed on 27 May 2023)) and are part of the Natura 2000 network (http://ec.europa.eu/environment/nature/natura2000/index_en.htm (accessed on 27 May 2023)). Most of the lagoons are traditional fishery fields, exploited as common extensive cultured systems based on the seasonal entrance of young fish into the lagoons and the autumn-to-winter offshore fish migration. In six lagoons, for each lagoon, the proportions of the annual fisheries landings to the total (during 1980–2008) landings per species ranged from 0.1 to 0.56 (mostly between 0.45 and 0.56) for mugilids, from 0.08 to 0.38 for European eels, from 0.09 to 0.58 (mostly between 0.09 and 0.15) for gilthead seabream, and from 0.02 to 0.15 for gobies (mainly Zosteriosessor ophiocephalus) [42].

In the Amvrakikos Gulf, the recorded annual wintering population of cormorants ranged from 1800 to 16,000 individuals during 2002–2022 (data provided by the Management Unit of Acheloos Valley and Amvrakikos Gulf Protected Areas and Hellenic Ornithological Society) using the midwinter census method [43]. Cormorants mainly roost on the small rocky island called Gaidaros (surface 5.6 × 10⁻³ km⁻²), located in the western part of the gulf (Figure 1) and their daily excursion for feeding extends up to a radius of 25 km (daily flight zone: DFZ) [31,32] (Figure 1).

Fisheries data consisted of the annual landings per species and lagoon during 1977–2020, provided by the Fishery Department of Preveza, Arta, and Lefkada island. Fisheries data for mugilids were disaggregated at three commercial categories (Cat), namely, “mugilids” Cat A (above 400 g), Cat B (between 150 and 400 g), and Cat C (between 80 and 150 g).

2.2. Modeling Approach

The basic concept of the model based on the fish size of mugilids that is consumed by the cormorants is smaller compared to the minimum length size of capture by fishers. Thus, the daily biomass (B_cr) of a fish species consumed by one individual of cormorant per day was estimated by

$$B_{cr}=DFI=N_{cr}·W_{cr}$$

where the DFI is the daily food intake of a cormorant, N_cr is the number of individuals of prey fish per cormorant, and W_cr is the mean weight of the fish species. The N_cr individuals are converted at the same length (L_cr) using the weight–length relationship, $W=a\cdot L^b$ (a and b the parameters of the length–weight relationships of fish species), and at W_cr and the age of preyed individuals (t_cr) using the length–age estimated by the von Bertalanffy equation, $L_t=L\infty \cdot (1-e^{-k(t-t_o)})$ (L∞, k, t_o are the parameters of the equation for each fish species).

When only the natural mortality existed (Μ), the number of individuals (N_c) at age t_c of exploitation for fish species was estimated by the following equation [42]:

$$N_c=N_{cr}\cdot e^{-\left(t_c-t_{cr}\right)·M}$$

The fishing biomass (Β_c) which corresponds to the fish species individuals that consumed by the cormorant per day is

$$B_c=catchR·N_c·W_c$$

where W_c is the mean weight of species at the fishing age (t_c), which is estimated by the equation $W_c=a·{\left(L\infty \cdot \left(1-e^{-k\left(t_c-t_o\right)}\right)\right)}^b$, and catchR is the catch rate.

The economic losses (E_c) are $E_c=B_c·v_j$, where v is the economic value (€) on each weight category (Cat) j (mean weight> 400 g: Cat A, 150 ≤ mean weight ≤ 400 g: Cat B, and 80 ≤ mean weight ≤ 150 g: Cat C).

In the case of the inclusion of i fish species in the predator’s trophic spectrum, the number of individuals of species i consumed by one cormorant (N_cri) is

$$N_{c{r}_i}=\frac{B_{cr}}{W_{{cr}_i}}·q_i$$

(1)

where q is the biomass proportion of species i in daily consumed biomass.

On a daily basis, the losses in fishing production (tB_c), the proportion of tB_c per weight category j (P_rj), the economic losses (tE_c), the average time that the losses will appear after the consumption of a fish (aT), and the average length of a consumed fish (aL_cr) per each individual cormorant were estimated, respectively:

$${tB}_c=\sum _{j=1}^3\sum _{i=1}^5\left(B_{c_{t_{j,i}}}\right)$$

$${Pr}_j=\sum _{i=1}^5\left(\frac{B_{c_{i,j}}}{{tB}_c}\right)$$

$$t{E}_c=\sum _{j=1}^3\sum _{i=1}^5\left(B_{c_{t_{j,i}}}·v_j\right)$$

$$aT=\sum _{i=1}^5\frac{\left(N_{c_i}·\left(t_{c_i}\right.-\left.t_{{cr}_i}\right)\right)}{\sum _{i=1}^5\left(N_{c_i}\right)}$$

$${aL}_{cr}=\sum _{i=1}^5\frac{\left(N_{{cr}_i}·L_{{cr}_i}\right)}{\sum _{i=1}^5\left(N_{{cr}_i}\right)}$$

2.3. Model Calibration and Sensitivity Analysis

The model was calibrated using the estimated biological parameters found in the literature (

Table A1. von Bertalanffy’s equation coefficients, length–weight equation coefficients, natural mortality (M), marker values (v) per weight category, daily food intake of birds (DFI), length of fish consumed (Lcr).

		Mugilids Species
		Chelon saliens	Chelon aurata	Mugil cephalus	Chelon ramada	Chelon labrosus
Total length	L∞ (cm)	32.99	69.30	79.10	56.33	47.20
Weight	W∞ (gr)	299.77	2656.31	4960.19	1873.70	1128.28
	k	0.261	0.136	0.151	0.179	0.200
	t0	−0.470	−1.140	−0.100	−0.856	−0.400
W = a × TLb	a	0.00784	0.0056	0.0072	0.0055	0.008
	b	3.018	3.130	3.041	3.160	3.010
	Reference	[43]				[50]
	Tm (°C)	18					Reference
	M #	0.587	0.312	0.322	0.396	0.448	[51]
	weight category	Cat A	Cat B	Cat C
	Weight range	>400 gr	150–400 gr	80–150 gr			[48]
	v (€.kg−1)	[24]	[51]	[43]			market values 2015–2020
(DFI) Daily food intake of Birds,	DFI for mugilids (gr)	169.6					[24]
(Lcr) Length of fish consumed	Lcr for mugilids (cm)	15–24					[24]
	q	0.134	0.089	0.147	0.341	0.293	[24]

# Log (M) = −0.0066 − 0.279 log (L_∞) + 0.6543 log (K) + 0.4634 log (Tm) ± 0.245 [51].

, Appendix A) and received random values around the M_i (Table A1) from a normal distribution (average value = 1; SD = 0.1) (Mcg), for the q_i from a uniform distribution at range [0.08–0.40], for the Lcr_i from a normal distribution (average value = 19.3 cm; SD = 1.3 cm; Table A1), for the DFI from a normal distribution (average value = 0.18 kg: SD = 0.04 kg) and for the catchR from a uniform distribution at range [0.2–1.0]. At the time of capture (t_c), the model received random values according to three cases: (1) from a beta distribution (α = 1, β = 3) at a range of integer values from 1 to 6, (2) from a uniform distribution at a range of integer values from 1 to 6; and (3) from a symmetrical distribution of the case (1) at a range of integer values from 1 to 6. Three different fishing strategies were followed based on targeted fish size: all fish sizes (1), non-targeted fishing (2), and targeted fishing of the larger fish sizes (3).

For each fishing strategy, the model was iterated 20,000 times, and in each iteration, the variables and the estimates of the model were recorded (

Table A2. Descriptive statistics of input and output of model data for 60,000 iterations.

No	Species	Situation	Variable	abr	Mean	SD	Min	Max
Input data
1	Chelon saliens	consumed	W (gr)	CS Wcr	36.30	12.03	4.02	100.22	estimated by #17 and #6
2	Chelon aurata		W (gr)	CA Wcr	36.41	12.07	4.74	98.40	estimated by #17 and #7
3	Mugil cephalus		W (gr)	MC Wcr	36.38	12.05	3.41	106.27	estimated by #17 and #8
4	Chelon ramada		W (gr)	CR Wcr	36.40	12.02	4.73	96.57	estimated by #17 and #9
5	Chelon labrosus		W (gr)	CL Wcr	36.40	12.04	5.31	98.09	estimated by #17 and #10
6	Chelon saliens		W Proportion	CS q	0.20	0.05	0.08	0.40	random distribution
7	Chelon aurata		W Proportion	CA q	0.20	0.05	0.08	0.40	random distribution
8	Mugil cephalus		W Proportion	MC q	0.20	0.05	0.08	0.40	random distribution
9	Chelon ramada		W Proportion	CR q	0.20	0.05	0.08	0.40	random distribution
10	Chelon labrosus		W Proportion	CL q	0.20	0.05	0.08	0.40	random distribution
11	Chelon saliens		L (cm)	CS Lcr	19.30	1.30	14.25	25.10	normal distribution
12	Chelon aurata		L (cm)	CA Lcr	19.30	1.31	13.79	25.07	normal distribution
13	Mugil cephalus		L (cm)	MC Lcr	19.30	1.30	13.94	24.93	normal distribution
14	Chelon ramada		L (cm)	CR Lcr	19.31	1.31	13.77	24.51	normal distribution
15	Chelon labrosus		L (cm)	CL Lcr	19.30	1.30	12.84	24.79	normal distribution
16			mean length (cm)	Lcr	19.09	0.63	16.10	21.69	estimated by #11 to #15
17			DFI (kg)	DFI	0.18	0.04	0.03	0.32	normal distribution
18			individuals	Ncr	3.12	0.68	0.60	6.80	estimated by #1 to #15
19	Chelon saliens		M range	CS M	0.59	0.06	0.34	0.83	estimated by Mi and #24
20	Chelon aurata		M range	CA M	0.31	0.03	0.18	0.44	estimated by Mi and #24
21	Mugil cephalus		M range	MC M	0.32	0.03	0.19	0.46	estimated by Mi and #24
22	Chelon ramada		M range	CR M	0.40	0.04	0.23	0.56	estimated by Mi and #24
23	Chelon labrosus		M range	CL M	0.45	0.04	0.26	0.64	estimated by Mi and #24
24			M fluctuation	Mcg	1.00	0.10	0.58	1.42	normal distribution
25			catchability (proportion)	catchP	0.60	0.23	0.20	1.00	random distribution
26	Chelon saliens	captured	Lc (cm)	CS Lc	26.87	2.87	18.66	31.31	beta distribution and random distribution
27	Chelon aurata		Lc (cm)	CA Lc	37.25	7.85	21.47	49.20	beta distribution and random distribution
28	Mugil cephalus		Lc (cm)	MC Lc	42.55	9.89	23.68	56.95	beta distribution and random distribution
29	Chelon ramada		Lc (cm)	CR Lc	35.52	6.68	21.09	45.27	beta distribution and random distribution
30	Chelon labrosus		Lc (cm)	CL Lc	32.31	5.25	19.97	40.00	beta distribution and random distribution
Output model
31	Chelon saliens	captured	W (gr)	CS Wc	9.85	8.93	0.20	78.55	estimated
32	Chelon aurata		W (gr)	CA Wc	55.55	33.37	4.07	341.29	estimated
33	Mugil cephalus		W (gr)	MC Wc	75.00	46.60	3.91	433.98	estimated
34	Chelon ramada		W (gr)	CR Wc	36.68	21.99	2.36	240.28	estimated
35	Chelon labrosus		W (gr)	CL Wc	22.31	14.48	0.88	135.13	estimated
36	Chelon saliens		E (Euro)	CS Ec	0.02	0.01	0.00	0.12	estimated
37	Chelon aurata		E (Euro)	CA Ec	0.25	0.20	0.00	2.05	estimated
38	Mugil cephalus		E (Euro)	MC Ec	0.39	0.30	0.01	2.60	estimated
39	Chelon ramada		E (Euro)	CR Ec	0.15	0.11	0.00	1.44	estimated
40	Chelon labrosus		E (Euro)	CL Ec	0.06	0.04	0.00	0.41	estimated
41			aT (year)	aT (year)	3.46	1.43	1.00	6.00	estimated
42			fishery biomass (kg)	Wc	0.20	0.10	0.02	0.76	estimated
43			E (Euro)	Ec	0.86	0.52	0.05	4.45	estimated
44			mean proportion	Cat A	0.50	0.33	0.00	1.00	estimated
45			mean proportion	Cat B	0.39	0.28	0.00	1.00	estimated
46			mean proportion	Cat C	0.11	0.16	0.00	1.00	estimated

, Appendix B). In each iteration, the Mcg was kept the same for all species, whereas the other random values were independent. For each fishing strategy, the distribution of the time that the losses would appear was also estimated.

A Kruskal–Wallis test (test-statistic; p = 0.05) was applied to check the significant differences found in tB_c and tE_c among the different fishing strategies, and a Mann–Whitney test (U; p = 0.05) was also used to identify the differences found in the mean values of tB_c and tE_c among the different fishing strategies [44].

Linear relationships among the log-transformed tB_c and tE_c with Mcg, aLcr, catchR, and Wcr_i were also applied based on a multi-regression model (MREG) for each fishing strategy:

$$Log\left(Vb\right)=c+b_1·Mcg+b_2·aLcr+b_3·catchR+\sum _{i=1}^5\left(k_i\right.·\left.W_{{cr}_i}\right)+SE$$

where V_b is tB_c or tE_c, the c, b’s, and k_i are coefficients estimated by the least squares regression techniques and SE is the standard error of the estimate. Significant variables used in the final model were selected using the backward stepwise variable selection method (F-to-remove; p ≤ 0.05) [44].

A sensitivity analysis provides an estimate of changes in the tB_c and tE_c values produced by the fluctuation of the above-stated parameters. This determines the parameters with a major influence on the tB_c, tE_c predictions. The analysis was conducted by means of successive simulations involving all parameters included in the model, varying by 20% above or below their initial baseline values (mean values of Mcg, aLcr, catchR, and Wcr_i).

2.4. Fishing Patterns in the Lagoons

A multivariate hierarchical cluster analysis based on the Bray–Curtis similarity index was also used to define similarities/dissimilarities of fisheries patterns among the studied lagoons using for each lagoon the mean (for the period 2002–2020) proportion of mugilid landings per each category and the mean proportion of simulated weight categories for each fishing strategy.

2.5. Distribution of Losses at Lagoons and Time Appearance of Losses

The total losses X (X = tB_c or TE_c) per lagoon s at a given year was estimated by

$$X_s=\frac{A_s}{\sum _{s=1}^s\left(A_s\right)}\times WP\times N_{bird}\times X_l$$

where A is the area of each lagoon, WP the wintering period (65 days) of cormorants in the study lagoons, N_bird the number of cormorants at a given year, and l the different fishing strategies.

The time distribution of appearance of losses after the predation time in the lagoon s is the same as the aT distribution of the fishing pattern that is used to classify the lagoon.

The quantity Xs per year (Xs_y) is calculated as ${Xs}_y=\sum _{d=1}^6\left({Fr}_{y+d}\right.·\left.{Xs}_{y-d-1}\right)$ where the y is the year of Xs estimation, Fr the proportion of X at years d after fishes preyed on birds.

The above model was developed in Excel (Microsoft Corporation, 2018), and the maps were created in QGIS (QGIS Development Team, 2009).

3. Results

The descriptive statistic of model variables and results are given in Table A2. The mean value of tE_c ranged from 0.614 (FP1) to 1.075 € bird⁻¹ day⁻¹ (FP3) and the mean value of tW_c ranged from 0.196 (FP1) to 0.203 kg bird⁻¹ day⁻¹ (FP2). Both variables (tE_c and tW_c) showed statistically significant differences among the fishing patterns (tE_c: FP3 > FP2 > FP1 and tW_c: FP3 = FP1 < FP2) (Kruskal–Wallis test; p < 0.05; Mann–Whitney test: U = 93.10⁸; p < 0.05) (

Table 1. Mean value and standard deviation (SD) of tE_c and tW_c and range of their values that correspond at the range of cumulative frequency (CF) from 0.025 to 0.975, per fishing pattern (FP). n is the number of model iterations, and the same letter marks the non-statistically significant mean values of groups (Mann–Whitney test: U; p > 0.05).

	tEc (€·bird−1·day−1)		tWc (kg·bird−1·day−1)
FP	Mean (SD)	CF Range (0.025–0.975)	Mean (SD)	CF Range (0.025–0.975)	n
1	0.614 a (0.376)	0.151–1.538	0.196 a (0.093)	0.061–0.399	20,000
2	0.886 b (0.499)	0.221–2.064	0.203 b (0.099)	0.059–0.424	20,000
3	1.075 c (0.554)	0.306–2.351	0.198 a (0.099)	0.057–0.429	20,000
Total	0.857 (0.518)	0.186–2.112	0.199 (0.097)	0.058–0.416	60,000

). The 95% of tE_c values ranged from 0.151 to 1.538 € bird⁻¹ day⁻¹ (FP1), from 0.221 to 2.064 € bird⁻¹ day⁻¹ (FP2), and from 0.306 to 2.351 € bird⁻¹ day⁻¹ (FP3). The 95% of tW_c values ranged from 0.061 to 0.399 kg bird⁻¹ day⁻¹ (FP1), from 0.059 to 0.424 kg bird⁻¹ day⁻¹ (FP2), and from 0.057 to 0.429 kg bird⁻¹ day⁻¹ (FP3) (Table 1).

The weight category composition differed among the different fishing strategies (χ² = 3214.9; df 4; p < 0.05). FP1 was mostly characterized by Cat B (58%), and FP2 by both Cat A (52%) and Cat B (39%), whereas FP3 was mostly characterized by a high percentage of Cat A (81%) (Figure 2).

The time appearance of economic and biomass losses after the preyed time differed among the fishing strategies (χ² > 55,167; Df = 10; p < 0.05) for tE_c and tW_c. For fishing strategy 1, 92% of economic and 94% of biomass losses appeared during the 1st and 2nd year after the impact. For fishing strategy 2, 76% of economic and 78% of biomass losses appeared during the 2nd and 3rd year after the impact, whereas for fishing strategy 3, 80% of economic and 79% of biomass losses appeared during the 4th and 5th year after the impact (

Table 2. Losses of tE_c (€·bird⁻¹·day⁻¹) and tW_c (kg·bird⁻¹·day⁻¹) at years (aT) after the impact per each different fishing strategy (FS). Brackets indicate the ratio and bold numbers indicate the highest values.

	FS	aT (Years)						Mean Value
		1	2	3	4	5	6
tEc	1	0.160 (0.26)	0.407 (0.66)	0.046 (0.07)	0.001 (0.00)	0.000 (0.00)	0.000 (0.00)	0.615
	2	0.017 (0.01)	0.317 (0.36)	0.350 (0.39)	0.155 (0.17)	0.034 (0.03)	0.002 (0.00)	0.879
	3	2.857 (2.65)	0.006 (0.00)	0.088 (0.08)	0.358 (0.33)	0.509 (0.47)	0.112 (0.10)	1.075
	total	0.059 (0.06)	0.243 (0.28)	0.162 (0.18)	0.171 (0.20)	0.181 (0.21)	0.038 (0.04)	0.856
tWc	1	0.064 (0.32)	0.121 (0.61)	0.010 (0.05)	0.000 (0.00)	0.001 (0.00)	0.000 (0.00)	0.197
	2	0.006 (0.03)	0.082 (0.41)	0.075 (0.37)	0.029 (0.14)	0.006 (0.03)	0.000 (0.00)	0.201
	3	9.922 (4.99)	0.001 (0.00)	0.018 (0.09)	0.068 (0.34)	0.090 (0.45)	0.019 (0.09)	0.198
	total	0.023 (0.11)	0.068 (0.34)	0.035 (0.17)	0.032 (0.16)	0.032 (0.16)	0.006 (0.03)	0.199

). The tE_c and tW_c exhibited a strong linear relationship (log(tE_c) = 1.476 + 1.043 × log(tW_c) ± 0.321; R² = 0.739; df:1.599; p < 0.05), and therefore, the multi-regression and sensitivity analysis applied only to the tE_c. At all fishing strategies the multi-regression analysis was significant (R² > 0.651; df:1,200; p < 0.05). The catchP and the fish species weight during consumption by a bird (Wcr_i) exhibited positive association with the tE_c, whereas the M_cg and L_cr showed a negative association with the tE_c.

Sensitivity analysis indicated that changes by ±20% in the tE_c exhibited changes from 0.48% (FP3: Wcr_LS) to 54.49% (FP3: Lcr). Between fish species, the highest change in the tE_c was shown by the M. cephalus (from 7.99% to 9.69%) and by the C. aurata (from 5% to 6.65%). Changes in catchP by ±20% exhibited changes in tE_c ranging from 19.9% to 24.9%. Changes in Mcg by ±20% exhibited changes in tE_c ranging from 15.89% to 18.89% for FP1, from 23.13% to 30.09% for FP2, and from 28.94% to 40.73% for FP3. Changes in aLcr by ±20% exhibited changes in tE_c ranging from 6.03% to 6.42% for FP1, from 27.04% to 37.07% for FP2, and from 35.27% to 54.49% for FP3 (

Table 3. Coefficients and standard error (in brackets) of multi-regression analysis among the natural log-transformed tEc and the independent variables (Int.Var), and sensitivity analysis per fishing strategy. N is the number of model iterations. Wcr is the mean weight of CS: Chelon saliens, CA: Chelon aurata, MC: Mugil cephalus, CR: Chelon ramada, CL: Chelon labrosus.

	Coefficients (Standard Error)			Sensitivity Analysis
	Fishing Strategy
	1	2	3	1	2	3
Constant (c)	−1.62 (0.0823)	0.449 (0.0704)	1.660 (0.0358)
Int. Var				%changes in tEc when Int.Var change ±20%
catchP	1.851 (0.0109)	1.863 (0.0093)	1.853 (0.0047)	19.92–24.88	20.04–25.06	19.93–24.90
Mcg	−0.860 (0.0254)	−1.31 (0.0213)	−1.700 (0.0109)	15.89–18.89	23.13–30.09	28.94–40.73
aLcr	−0.010 (0.0040)	−0.08 (0.0034)	−0.110 (0.0017)	6.03–6.42	27.04–37.07	35.27–54.49
WcrCS	0.001 (0.0002)	0.000 (0.0001)	0.000 (9.5223)	0.92–0.93	0.51–0.51	0.48–0.48
WcrCA	0.007 (0.0002)	0.007 (0.0001)	0.008 (0.0000)	5.00–5.26	5.57–5.90	6.24–6.65
WcrMC	0.011 (0.0002)	0.012 (0.0001)	0.012 (9.5908)	7.99–8.68	8.46–9.24	8.84–9.69
WcrCR	0.005 (0.0002)	0.005 (0.0001)	0.004 (9.6104)	4.25–4.44	4.04–4.21	3.47–3.60
WcrCL	0.002 (0.0002)	0.002 (0.0001)	0.001 (9.5399)	1.90–1.94	1.46–1.48	1.22–1.24
R2	0.651	0.734	0.918
n	20,000	20,000	20,000

).

The mean total yield of lagoons during 2002–2020 ranged from 1.40 (±1.01) t·km⁻² (No10, Tsopeli) to 4.40 (±3.59) t·km⁻² (#14, Vathi), whereas the mean yield of mugilids ranged from 0.11 (±0.09) t·km⁻² (No12, Pogonitsa) to 1.61 (±1.58) t·km⁻² (No14, Vathi). The mean proportion of mugilids to total lagoon landings was 0.44 and ranged from 0.05 (No12, Pogonitsa) to 0.55 (No3, Avlemonas). The corresponding estimates for Cat A ranged from 0.21 (#9, Logarou) to 1 (No6, Agrilos); for Cat B, they ranged from 0 (No6, AGR) to 0.59 (No9, Logarou); and for Cat C, they ranged from 0 (No6, Agrilos) to 0.35 (No3, Avlemonas).

Hierarchical clustering in the mean (for the period 2002–2020) proportion of mugilids landings per commercial category showed that (Figure 3 and

Table 4. Mean total yield (Yield T), mean yield of mugilids (Yield M), weight proportion of mugilids to total landings (PrpM), n the years by available data, mean proportion of weight categories of mugilids (Cat A, Cat B, and Cat C) and fishing strategy (FS) and estimated fishing strategy (FSe) by cluster analysis, for 2002–2020. No indicates the codes of the studied lagoons according to Figure 1. * indicates lagoons Nos 1 and 4, for which fishing strategy 2 was applied.

No	Area (km2)	Yield T (t·km−2)	Yield M (t·km−2)	PrpM	n	Cat A	Cat B	Cat C	FSe
1	2.26	Unknown fishing activity and fishing data during 2002–2020							2 *
2	3.29	1.30 (0.75)	0.48 (0.22)	0.37	19	0.42	0.29	0.29	2
3	5.25	1.36 (0.80)	0.76 (0.39)	0.55	14	0.37	0.28	0.35	2
4	0.66	Unknown fishing activity and fishing data during 2002–2020							2 *
5	0.56	Unknown fishing activity and fishing data
6	1.94	0.29 (0.15)	0.13 (0.10)	0.45	5	1.00	0.00	0.00	3
7	3.22	1.28 (0.78)	0.19 (0.16)	0.14	6	0.54	0.32	0.15	2
8	0.39	Unknown fishing activity and fishing data
9	30.64	2.65 (0.83)	1.24 (0.56)	0.46	16	0.21	0.59	0.20	1
10	1.12	1.40 (1.01)	0.65 (0.43)	0.46	19	0.75	0.20	0.05	3
11	1.90	2.29 (1.84)	0.60 (0.52)	0.26	18	0.74	0.16	0.10	3
12	0.47	2.39 (2.40)	0.11 (0.09)	0.05	12	0.58	0.42	0.00	2
13	43.90	0.73 (0.27)	0.36 (0.14)	0.48	17	0.31	0.49	0.20	1
14	0.29	4.40 (3.59)	1.61 (1.58)	0.36	18	0.41	0.47	0.13	1
				FS	1	0.18	0.59	0.24	1
					2	0.50	0.40	0.10	2
					3	0.81	0.19	0.01	3

), in a similarity index higher than 0.75, Logarou, Tsoukalio, and Vathi lagoons followed fishing strategy 1; Palaio, Avlemonas, Koftra-Palaiompouka, Pogonitsa, Saltini, and Myrtari lagoons followed fishing strategy 2; and Agrilos, Tsopeli, and Mazoma lagoons followed fishing strategy 3. In the Rouga and Sakolesi lagoons, no losses by birds were estimated, because we are not aware on the applied fishing strategy (Figure 3 and Table 4).

The time series of economic and biomass losses caused by the cormorants in the lagoons of the Amvrakikos Gulf from 2002 to 2020 are forecast up to 2026. Their picks follow the time series of bird numbers with a delay of about 2 years (Figure 3A,B). The mean annual number of cormorants was significantly (Mann–Whitney test: U = 12; p < 0.05) increased in 2002–2010 and 2011–2020. More specifically, the number of cormorants ranged from 1848 to 5823 bird·year⁻¹ (mean value of 3239.33 bird·year⁻¹) during the period 2002–2010 and from 4181 to 8375 bird·year⁻¹ (mean value of 5591.77 bird·year⁻¹, excluding an extreme value of 16,236 birds) in 2011–2020.

The estimated economic losses ranged from 0.04 to 0.35 MEuro·year⁻¹ (mean value of 0.12 MEuro·year⁻¹) during 2002–2010, from 0.07 to 0.70 MEuro·year⁻¹ (mean value of 0.24 MEuro·year⁻¹) during 2011–2020, and from 0.02 to 0.20 MEuro·year⁻¹ (mean value of 0.07 MEuro·year⁻¹) during 2021–2026 (Figure 3C). The estimated landing losses ranged from 12.49 to 82.71 t·year⁻¹ (mean value of 35.78 t·year⁻¹) during 2002–2010, from 25.88 to 172.04 t·year⁻¹ (mean value of 74.31 t·year⁻¹) during 2011–2020, and from 7.22 to 48.69 t·year⁻¹ (mean value of 20.89 t·year⁻¹) during 2021–2026 (Figure 3C). Six lagoons (Figure 1: No13, Tsoukalio-Rodia; No9, Logarou; No3, Avlemonas; No2, Palaio; No7, Koftra-Palaiompouka; and No6, Agrilos) cumulatively contributed to the 89.7% and the 92.8% of the total tE_c and tW_c, respectively, with the first two lagoons (Tsoukalio-Rodia and Logarou) exhibiting 70.2% and 78.1% of the total tE_c and tW_c, respectively (Figure 3D).

In the Tsoukalio-Rodia lagoon (Figure 1: No13), the sum of the biomass losses during 2002–2020 was 446.6 t (ranging from 157.87 to 1024.82 t), and in Logarou (Figure 1: No9), it was 311.7 t (ranging from 110.19 t to 715.33 t). The sum of landings was 269.6 t and 608.6 t for Tsoukalio-Rodia and Logarou (Figure 1: Nos 13 and 9), respectively (Figure 3E₁), and the ratio of biomass losses to landings biomass ranged from 0.58:1 to 3.80:1 and from 0.18:1 to 1.17:1, respectively. The ratio of economic losses to income from landings ranged from 0.41:1 to 4.18:1 and from 0.14:1 to 1.42:1, respectively (Figure 3E₁,E₂).

4. Discussion

In the present study, an attempt was made through modeling to estimate the losses of lagoon fisheries from predation of cormorants during the wintering period. A substantial number of uncertainties have been incorporated into the model regarding the fish population growth, fish size and species distribution, age (or size) of fish that might be caught, and catchability as well as the different fishing strategies followed by the lagoon fishers. On the other hand, the large number of the model’s iterations reduced the uncertainties of a large number of the aforementioned parameters.

Mugilids are the dominant fish group in the Greek lagoon fisheries [6] (>50%), and they also comprise the dominant prey of cormorants during their wintering period (74% of DFI in Mesolongi-Aitoliko lagoons: [45]; 65% of DFI in Amvrakikos Gulf lagoons: [24]). The range of DFI for mugilids in the model agrees with previous work in the study area [24]. The length of preyed fish (except for C. saliens) ranged from 15 to 24 cm [45], supporting the basic assumption that the losses induced by the impact of cormorants would appear in future [41].

The model assumed (a) that natural mortality was constant during the years, (b) that preyed fish for each species were of the same size, (c) spatial homogeneity of species composition among the lagoons, (d) that mean preyed fish species composition was constant day to day as well as during the years, (e) spatial distribution of the cormorants, and (f) that the wintering period of the cormorants was defined as 65 days.

(a)

The natural mortality M was ±42% in each species, which might include the inter-annual variability and the variability driven by the von Bertalanffy estimates (L_∞ and k participate in the M estimation; Table A1).

(b)

The maximum number of preyed individuals was approximately 7 (Table A2). Given that mugilids exhibit a schooling behavior according to the individuals’ size [46], the likelihood that individuals of the same species in a feeding area for cormorants are the same size was high.

(c)

The studied lagoons exhibited differences in physicochemical variables [42], which explains the spatial expansion of mugilids according to their preferences [46]. Model iterations recorded a series of cases that simulated variable species composition. In each case, this affects the average value of the model’s estimates but not the range of distributions.

(d)

The analysis indicated that most of the losses are predicted to be in the two largest lagoons (No 13, Tsoukalio-Rodia and No 9, Logarou; Figure 1). In these lagoons, the high representation of mugilids is maintained at the same levels [42]. According to fishers, most of the losses in the above-mentioned lagoons were on the gilthead seabream. However, the gilthead seabream (except for a small lagoon: No 12, Pogonitsa; Figure 1) consists of a relatively small proportion to landings (<15%), and in some cases, the production was supported by enrichment programs [42]. It seems that during periods of low temperatures, gilthead seabream searches for favorable sites and is concentrated near to the communication, with the sea channels being an easy target for cormorants. In each case, this impact is temporarily limited to a few days, and it might not be sufficient to alter the prey species’ composition in the study lagoons.

(e)

The bird-days establishment of the cormorants in each lagoon ranged from 12 to 110 bird·days·year⁻¹·ha⁻¹, which is in agreement with the estimates reported by a previous study [13] (inland waters/lagoons: 20 to 100 bird·days·year⁻¹·ha⁻¹). Certain wetlands have been excluded from our estimations (e.g., the Amvrakikos Gulf, rivers Louros and Arachthos, Lake Voulkaria). The area of the gulf is apparently used by cormorants in limited situations for feeding, such as preying on small pelagics, which consists of 2.8% of DFI. Given also that freshwater fish species are not included in the feeding spectra of cormorants [24], the adjacent lakes and the upper system of rivers have not been considered in this study.

(f)

The wintering population of cormorants in the studied area referred to the maximum number of midwinter estimates. However, the first appearance of the cormorants in the Mediterranean is observed from the mid October and the last ones in mid April, with a progressive increase during the winter and a gradual decrease after the maximum appearance [47]. Through the use of a normal distribution, the timing of the appearance was simulated (mean = 90 days, SD = 25.7 days) and indicated that the recorded bird value (maximum value of distribution) multiplied by 64.39 days was equal to the bird-days estimated from the simulated appearance of cormorants. Thus, the 65 days estimated as the wintering period could be considered as a reliable estimate.

The model’s sensitivity analysis revealed that the Mcg and the aLcr mostly affected the losses, according to fishing strategy. The most sensitive scenario was FP3 (fishing targeting greater-size fishes). In the case of high natural mortality, some of the individual fish were not caught in this scenario, so low values of M indicate greater losses than high values of M, which indicate smaller losses. The aLcr also affected the losses. Greater losses than the preyed aLcr are suggested when the preyed occurs at low length. This is also expected due to the fact that when the aLcr is low, a higher preyed mortality (consumed more individuals) was estimated rather than when the aLcr is high (consumed fewer individuals, and thus lower preyed mortality). In relation to fish species, sensitivity analysis indicated that the losses are more sensitive to the preyed flathead grey mullets, but the effect on losses was rather low (≈8 to 9.7%). It should be noted that for all fish species, the impact of cormorants referred to the prospective fishing biomass. From flathead grey mullet, a boutarga-type product called “avgotaracho” (dried, salted, and waxed ripe ovaries) is made, which corresponds to 7% of fishing biomass of the species [48,49]. Thus, the economic losses incurred by Cat A for flathead grey mullet were underestimated by 130%, revealing an important effect on losses for this species.

The ratio of biomass losses to landing biomass and the ratio of economic losses to incomes from landings differed between the two largest lagoons, supporting different measures against the impact of cormorants. The ratios of biomass losses to landings biomass and the ratio of economic losses to income from landings that ranged from 0.58:1 to 3.80:1 and from 0.18:1 to 1.17:1 (biomass) and from 0.41:1 to 4.18:1 and from 0.14:1 to 1.42:1 (economic) support a strong competitive relationship between cormorants and fisheries in the lagoons of the Amvrakikos Gulf. It seems that the yearly losses in biomass by cormorants is sufficient to justify this detrend of production. Finally, it is worth noting that the estimates in this study applied to the direct losses by cormorants and were underestimated due to the fact that, during the feeding, significant numbers of fish are injured [13,38] (mainly of commercial size), most of which die shortly after or survive and are later caught by fishers and sold at low value or discarded.