Methods

[Methods anthropogenic modifications]
[Methods flora]
[Methods invertebrates and fishes]
[Methods amphibians and reptiles]
[Methods birds]
[Methods mammals]
[Squares in study area]

Structure of aquatic databases

Surveys

The aquatic databases have the following overall structure:

The arrow indicates that there are generally several stations per documentary source, several types of gear per station and several taxa per gear type. Although this structure reflects the nature of sampling in an aquatic environment, it had to be changed to analyze biodiversity. The desired structure is:

A geographic entity refers to any square or region of the river created by a division of the territory by criteria such as depth or sediment type, or by a purely spatial division not based on environmental factors. For aquatic ecosystems, the divisions retained comprised:

Square grids

Using a grid to divide a study area into squares is the traditional method for creating divisions, since it makes it easier to create contiguous squares. Although squares of 10 km x 10 km were used for terrestrial data since this was found to be the best scale for most terrestrial survey data, preliminary tests showed that this resolution was too fine for a marine environment. Therefore, a grid of 20 km x 20 km squares was used for commercial dragging and trawling data. In the freshwater environment, however, 10 km x 10 km squares were found to be too large and tessellation was used instead.

Tessellation

Tessellation consists in dividing a territory into irregular polygons (called Thiessen or Voronoi polygons). Since the polygons follow the complex contours of the shoreline, they provide complete coverage of a study area, avoiding the main problem involved in using square grids: squares that are ill fitted to the territory under study. Localized survey stations, chosen at random, were used as the centres for constructing Thiessen polygons. Polygons were generated with ArcView software using a program by Ammon (1998). Readers are invited to consult Okabe et al. (1992) and Gold (1991) for more details on the properties and applications of Thiessen polygons. Survey stations were located in Thiessen polygons to obtain local lists of species complete enough to analyze biodiversity.

Status of taxa

Both freshwater and saltwater fishes were dealt with on the species level. The identification of invertebrates, however, often had to be done at a less precise taxonomic level, particularly in the case of insect larvae, which can often only be identified to genus. Particularly difficult or poorly known groups such as mites and nematodes were identified to class, order or family level. In the spatial analyses, we retained species records for marine invertebrates and freshwater zooplankton larvae and genus records for freshwater benthos.

After the necessary taxonomic revision, a brief review was conducted of the geographic distribution, ecology and conservation status of the taxa. The information was obtained mainly from syntheses on the ecology of aquatic organisms. Some of these works contained specific information at the species level (fishes, freshwater molluscs) or genus level (freshwater insects). However, for the other taxa, the works consulted often went no farther than the family or order level, with a few records on species. Information was also taken from monographs dealing with a number of different taxa. We did not carry out an exhaustive literature search of monographs to find species information, since this would have greatly exceeded the scope of the project.

Taxonomy

Integrating a number of lists of species into a coherent database required an exhaustive review of each taxon in order to uniformly apply taxonomic revisions and correct errors. Generally, we first drew up a list of all the names of the basic taxa, with the correct name and a code to facilitate further processing. For example, records for the rainbow trout might include:

A 16-character code was used to ensure the uniqueness of codes for freshwater and marine vertebrates and invertebrates, to guarantee readability and to provide for an increased database search capacity. In the example above (ch.sp.Onchorh,myk), ch is the abbreviation for vertebrates (chordés); sp for species; Oncorh for Oncorhynchus; and myk for mykiss.

The overall taxonomic revision was done with the ITIS database (1999). Other basic references included Coad (1995), Scott and Crossman (1974) and Scott and Scott (1988) for fishes; Brunel, Bossé and Lamarche (1998) for marine invertebrates and Merritt and Cummins (1996) and Brusca and Brusca (1990) for freshwater invertebrates.

Distribution

Distribution was described at two levels. For worldwide distribution, taxa were divided into broad categories (cosmopolitan, circumboreal, amphi-Atlantic or endemic to North America), emphasizing endemic North American species. For North American distribution, distribution was specified by Canadian province or U.S. state for freshwater organisms and by broad coastal marine zones for marine organisms (Arctic, Labradorean, Acadian, Virginian, Carolinian and Caribbean).

Breeding status

The breeding status indicates whether the species breeds in the study area.

Conservation status

The conservation status is based on the priority assigned to species by various committees of experts, compiled by the Quebec Department of the Environment and Wildlife (CDPNQ 1999).

Ecology

A literature review of general works on the ecology of St. Lawrence species shows that many species, particularly invertebrates, are poorly known. In many cases, we had to rely on the characteristics of the genus or family. Data were compiled on the following descriptors:

The most information was available on freshwater fishes (Coad 1995; Scott and Crossman 1974). For subsequent analyses, a score (0 to 10, for a possible total of 10) was awarded for the following descriptors:

Scores were also awarded for salinity tolerance (stenohaline, euryhaline), life cycle (freshwater, diadromous), maximum size attained and age at sexual maturity. The spawning substrate and time of spawning were also compiled. Data on ecophysiological constraints, such as tolerance of turbidity and extreme and optimal temperature ranges for growth and reproduction, were often scarce, however.

Similar criteria were employed to characterize marine fishes, and each species' preferred depth.

Freshwater invertebrates were classified according to two criteria, life style (benthic or pelagic) and feeding strategy (grazers, collectors, deposit feeders, shredders, parasites, predators, suspension feeders and borers). Marine invertebrates were classified mainly according to their size, life style (benthic or pelagic), tolerance of temperature and salinity variations and preferred range of depths.

Statistical analyses

Statistical analyses were carried out at a number of different spatial scales. The following list gives an idea of the type of data processed and the resulting analyses:

Scale of the St. Lawrence

Rarity

Saltwater fishes. – Survey trawls provide a fairly accurate picture of the frequency and abundance of marine species in trawlable areas, as long as the selectivity of the gear is taken into account. On the scale of the entire marine part of the St. Lawrence, each species was characterized by two criteria: frequency of occurrence (number of trawl tows in which the species was found) and mean abundance (average number of individuals by tow, when the species was observed). Species were grouped according to these two criteria, which were used simultaneously in group average clustering to determine the Euclidian distance between species.

Freshwater fishes. – Owing to the use of many different gear types, often poorly described, it was difficult to rank species by frequency and abundance, since the ranking obtained depends largely on the selectivity of the fishing gear used. Since no gear type covers all species, an alternative approach based on the use of a wide variety of gear types was adopted.

First, we determined the type of gear used to catch each species. (Most abundant or common species should be able to be caught by a range of gear types; the inverse should be true of rare species.) Then, the capture frequency for each species was determined for each type of gear and transformed into a rank value; the highest rank value obtained by a species was used to characterize it. This means that a species that is common but that was detected with only one, very selective, gear type will receive a high score, while a rare species will generally receive a low score for all gear types. Species were then classified according to these two criteria, used simultaneously in group average clustering to determine the Euclidian distance between species.

Regional species-area curve

There is generally a strong correlation between the size of an area and its species richness (Connor and McCoy 1979; McGuinness 1984; Levin 1996), which must be taken into account when comparing diversity between one or more units. The relationship between species richness and area was modeled using the logistical model of He and Legendre (1996):

Parameters B, C and Z were estimated for a set of values associating the observed richness S with a given area A. The equation was calculated for all of the study area by combining Thiessen polygons at random to form increasingly larger areas. This approach generates a species-area curve that is independent of environmental variables, such as location along the watercourse, depth or sediment type. The equation serves as a neutral model for comparing the effect of the variations summarized in the ecological divisions. Species-area curves were calculated separately for freshwater benthos, freshwater fishes and saltwater fishes.

Species-area curve and distribution of species richness (freshwater benthos)

The species-area curve for freshwater benthos was calculated as follows:

Readers should note that this relationship is exploratory and must be confirmed with additional data providing better spatial coverage.

Species-area curve and distribution of species richness (freshwater fishes)

The species-area curve for freshwater fishes was calculated as follows:

Species-area curve and distribution of species richness (marine fishes)

The species-area curve for marine fishes was calculated as follows:

The equation estimates the change in species richness S as a function of the area surveyed and serves as a basic model for evaluating the diversity of fishes in the marine St. Lawrence.

Scale of units

Intraunit analyses

The species richness (S) observed in a unit is largely dependent on the sampling effort (n). To determine if the observed species richness was close to the actual species richness in each unit (S_max), the relationship between species richness and effort was modeled using the following equation (Colwell and Coddington 1995):

where S_max and B are parameters estimated from the cumulative richness S after n samples. The parameter Smax represents the theoretical species richness in the unit, based on the accumulation of new species as n increases. The parameter B is interpreted as the inverse of the detectability of the species. As the following table shows, at B = 5, for example, it is expected that 50% of species will be caught after five samples.

At B = 5,

Since the estimation of S_max and B by nonlinear regression is influenced by the order in which samples are added, 100 iterations of the estimates were carried out, randomizing the order of samples. In subsequent treatments, the median of S_max was used to describe the central tendency of S_max, and the 5th and 95th percentiles, for dispersion. Estimates were rejected if convergence did not occur (formation of a single mode for values of S_max).

Interunit analyses

Cluster analyses

A cluster analysis technique was used to categorize units based on the similarity of the species they contained, to reveal the main clusters as well as existing discontinuities (Legendre and Legendre 1998).

The matrix [units-species occurrence] derived for the units retained was used to calculate a similarity matrix using the Jaccard coefficient, which was then employed in group average clustering (UPGMA).

The significance of the clusters obtained was verified with the permutation method described by Manly (1993). The statistic of interest in the test is the variance in the number of co-occurrences. If two units belong to two ecologically significant groups, they should share fewer species than predicted by the null hypothesis (random co-occurrences that only depend on the number of species in each unit). By constructing pseudo-matrices based on the null hypothesis a large number of times (100), one can test the variances of the co-occurrences in the true matrix, which should be higher than in most of the pseudo-variances.

No completely satisfactory method has been found for defining the number of recognized clusters (Panel on Discriminant Analysis, Classification and Clustering, 1998). In practice, this number depends basically on the ecological interpretation of the results.

Selectivity of species

To better gauge the ecological implications of the clusters obtained, a selectivity analysis can be useful for species that have been well sampled. Groups are characterized by the species that favour or avoid areas, or which appear to be neutral (non significant test). The groups of geographic units are then compared by using the traits of species that favour them, such as salinity tolerance, degree of association with vegetation and mobility.

The selectivity of species in relation to a given classification is provided by the formula (Neu et al. 1974):

which gives the confidence interval for the proportion p_i of records in the class i. The expected value for the proportion is obtained from the proportion of samples in class i. If the expected value is below the lower limit of the confidence interval (with a = 0.10 and the normal deviation Z corrected for k classes), the class is favoured by the species. If the expected value is greater than the upper limit of the interval, the class is avoided. If the expected value falls within the interval, the test is not significant.

The issue of taking account of fishing gear in estimations of local species richness

Introduction

The list of fish species in a large river is usually compiled from a number of different surveys carried out with different types of fishing gear. This type of meta-analysis of different databases is often recommended to increase the detection of species of fish or other animals (Rodda 1993; Boulinier et al. 1998; Kidric-Brown and Brown 1993; Gibbons et al. 1997). This approach is the inverse of specific surveys with a single, selective type of fishing gear (FAO 1975), which provides quantitative data on only some of the species present. In this study, first we grouped together gear types that generally detected the same species. Then, we tested the hypothesis that lists of species obtained without taking into account the gear type were more exhaustive than lists prepared with data from specific gear types, as well as the hypothesis that the former provided better spatial coverage.

Databases

The database on freshwater fishes in the St. Lawrence River consists of nearly 14,000 samples obtained between 1928 and 1997, most dating from between 1960 and 1997. The samples, georeferenced by longitude and latitude, can be grouped into geographic units (polygons) of variable area associated with a list of species by gear type and an overall list of species regardless of gear type. These lists do not take account of the potential temporal variations that could have occurred during the period covered.

Fishing gear diversity

The information on fishing effort and the type of fishing gear was often scarce or incomplete. Therefore, gear types were classified based on how well they were described and the frequency of use, but not according to fishing effort, for which the data were not complete enough. The database contains records of 39 types of fishing gear that detected a total of 102 fish species. This result is very similar to the species richness found in the St. Lawrence by Underhill (1986), who listed 98 species. Consequently, it can be concluded that, when taken together, the 39 types of gear describe the regional species richness of fishes in the St. Lawrence.

The fishing gear in the database can be divided into three broad categories:

Probability of detection

The first step consisted in identifying well-described gear types that caught more or less the same species, by calculating the probability of detecting species for the main gear types. To do this, stations were grouped into 1 km² squares. For each gear type, we calculated the number of squares where the gear was used and where the species was present, since the species was detected by one of the gear types used in the squares. The probability of detecting a species with a given gear type was calculated as:

Here is an example of how the detection probability is calculated for yellow perch (PEFL) with a 38-mm gillnet (stretched mesh):

The species detection-gear probability matrix was therefore calculated at the 1 km2 scale. For statistical reasons, calculations were done when the number of the squares where the species was present exceeded 30 (Cochran 1977).

The next step consisted in calculating correlations between pairs of detection probabilities, and using this to identify equivalent gear types. The following table shows the paired correlations between detection probabilities.

Correlation of probabilities of detection

	Beach seine	Electro- fishing	Line	Gillnet 25 mm	Gillnet 38 mm	Gillnet 51 mm	Gillnet 64 mm	Gillnet 76 mm	Gillnet 102 mm	Gillnet 127 mm	Gillnet 152 mm
Beach seine	1,0
Electro- fishing	0,8	1,0
Line	-0,1	0,0	1,0
Gillnet 25 mm	0,7	0,2	0,0	1,0
Gillnet 38 mm	-0,2	0,2	0,6	-0,3	1,0
Gillnet 51 mm	-0,5	0,0	0,5	-0,4	0,9	1,0
Gillnet 64 mm	-0,1	-0,4	0,7	-0,3	0,8	0,9	1,0
Gillnet 76 mm	-0,6	0,0	0,6	-0,6	0,9	0,9	0,7	1,0
Gillnet 102 mm	-0,7	-0,1	0,3	-0,7	0,9	0,7	0,3	0,9	1,0
Gillnet 127 mm	-0,8	-0,5	0,0	-0,7	-0,1	0,0	0,0	0,3	0,8	1,0
Gillnet 152 mm	-0,9	-0,6	0,1	-0,6	0,0	0,1	0,0	0,3	0,6	0,9	1,0

The correlation structure allowed gear types to be combined to analyze the local occurrence of species. The following groups were proposed:

     - Beach seine type (beach seine, gillnets with a mesh size of 25 mm and electrofishing);
     - 38-mm gillnet type (gillnets with a mesh size of 38, 51, 64 or 76 mm);
     - 102-mm gillnet type (gillnets with a mesh size of 102, 127 or 152 mm).

The beach seine type detected 91 out of 102 species, far more than the 38-mm gillnet type (54 species) and 102-mm gillnet type (41 species). The group of unspecified gear types represented 3279 samples and detected 98 species.

Comparison of lists of species

The division of the study area by tessellation was used to compare estimates of Smax (theoretical species richness of each unit) based on three groups of samples: those obtained with the beach seine gear type, those obtained with 38-mm gillnets and all samples regardless of gear type. Samples obtained with 102-mm gillnets were eliminated since this gear type is relatively uncommon and detects very few species.

Samples obtained with beach seine gear type

As the figure shows, the estimates were strongly positively correlated. However, it must be remembered that samples obtained with the beach seine gear type are a subset of the set of all samples obtained regardless of gear type. The scatter of the estimates of species richness along the diagonal line represents the additional variation resulting from the addition of gear types other than the beach seine type.

Samples obtained with the 38-mm gillnet

Similarly, samples obtained with 38-mm gillnets were compared with those obtained with all (unspecified) gear types, showing a strong positive correlation. The comments on method made in the previous paragraph also apply in this case.

These results show that a range of unspecified or poorly described gear types can be used to estimate local species richness if this results in a substantial gain in spatial coverage.

Spatial coverage

The use of all samples regardless of gear type provides good spatial coverage, as the following table shows. By limiting the analysis to the beach seine type, we would only obtain reliable species richness estimates in 120 tessellation units covering 91 species. As its name indicates, the beach seine can only be used near the shore and not in deep water. Gillnets with 38-mm mesh are more versatile in this respect, providing reliable estimates in 167 units; however, they are much more selective, detecting only 54 species. Samples from all gear types provided coverage of 247 units and 98 species. Four other species were cited in the database (to make the total of 102 species) but were not georeferenced.

Conclusion

The list of 102 species in the freshwater part of the St. Lawrence was compiled from nearly 14,000 samples obtained by 39 different kinds of fishing gear. Despite the large number of samples, no single type of gear was able to detect all the species. The most important conclusion to be drawn from our calculations is there is a strong correlation between species richness detected by specific gear types and that compiled without taking account of gear type. Therefore, the second option was retained, since the analyses were based on the occurrence of species in geographical units rather than on abundance. Due to this approach, we were able to take advantage of the large number of samples for which the gear type was unspecified or poorly described, which provided significantly better spatial coverage.



Jacques Leclerc