| Chapitre 1 Regional influence of mixedwood forest on bird distribution in Québec and Ontario | ||
|---|---|---|
|
|
|
| Chapitre 1 Regional influence of mixedwood forest on bird distribution in Québec and Ontario | ||
|---|---|---|
|
|
|
|
Table des matières
Ce chapitre sera éventuellement soumis à la revue «Landscape Ecology» avec les co-auteurs suivants : Caroline Girard, Charles Francis, Marcel Darveau et André Desrochers.
Nous avons utilisé les données des Atlas des oiseaux nicheurs du Québec et de l’Ontario et des images du satellite Landsat pour étudier, à une échelle régionale, l’utilisation de la forêt mixte par les oiseaux forestiers. La présence de 25 espèces sur 71 était distinctement influencée positivement par la forêt mixte alors que la présence d’une espèce, le durbec des sapins, présentait une relation négative avec ce type de couvert. D’autres espèces étaient associées aux forêts de conifères ou de feuillus ou à plus d’un couvert forestier. La fraction de variance expliquée par le couvert forestier était faible, d’autres facteurs, tant écologiques que méthodologiques, influçaient la présence des espèces dans les parcelles. Nous concluons que la forêt mixte possède une avifaune qui lui est propre et qu’elle ne constitue pas seulement un écotone présentant à la fois les espèces associées aux forêts boréales de conifères et aux forêts tempérées de feuillus.
We used the data of the breeding bird atlas of Québec and Ontario and imagery of Landsat satellite to study, on a regional scale, forest bird use of mixedwood forest. The occurrence of 25 species over 71 was distinctively positively influenced by mixedwood forest, whereas one species, pine grosbeak, was defined by a negative relation with this cover type. Other species were associated to coniferous, or decidous forest or to more than one forest cover. The variance fraction explained by forest cover was low, other factors, ecological as well as methodological, had influence on species presence. We conclude that mixedwood forests possess their own avifauna and that they are not only an ecotone that present bird species associated to coniferous boreal forests and deciduous temperate forests.
Mixedwood forests, “trees belonging to either of the botanical groups Gymnospermae or Angiospermae and which are substantially intermingled in stands” (Côté 2000), represent in Québec and in Ontario approximately 20% of the forests. Patchily distributed through a large strip that crosses the two provinces between the 43rd and the 49th parallels, they form a transition zone between boreal coniferous forests and temperate deciduous forests, and are therefore subdivided in two sub-regions, the temperate mixedwood in the South and the boreal mixedwood in the North.
Many species are known or suspected to use North American mixedwood forest, such as arthropods (Coleoptera, Diptera, Hymenoptera, and Acari) (Hammond 1997), mammals (squirrels, voles, mice, bats, black bear ( Ursus americanus ), American marten ( Martes americana ), fisher ( M. pennanti )) (Baker et al. 1995, Bayne and Hobson 1998, Crampton and Barclay 1998, Kalcounis et al. 1999), birds (barred owl ( Strix varia ), red-breasted nuthatch ( Sitta canadensis ), Swainson’s thrush ( Catharus ustulatus ), Tennessee warbler ( Vermivora peregrina ) and winter wren ( Troglodytes troglodytes )) (Norton and Hannon 1997, Mazur et al. 1998, Hobson and Bayne 2000, Schieck et al. 2000, Robichaud et al. 2002). Despite that fact, few studies have addressed directly the influence of mixed coniferous-deciduous forests on wildlife distribution. In western and northern Canada, Kirk et al. (1996) showed that some bird species (e.g. red-eyed vireo ( Vireo olivaceus ) and black-throated green warbler ( Dendroica virens )) occurred more frequently in boreal mixedwood stands than in deciduous or coniferous stands. Similarly, Hobson and Bayne (2000) found in Saskatchewan (Canada) a greater avian diversity and a higher abundance of certain species in boreal mixedwood stands than in “pure” deciduous or coniferous stands. Another study investigated the question at a broader scale ; Enoksson et al. (1995) showed the importance of the presence of deciduous patches within a coniferous forest landscape of Sweden for several bird species.
It is now well recognized in ecology that habitat selection is a process that takes place at different spatial scales and that variables influencing certain processes at one scale do not necessarily influence these processes at other scales (Cody 1985, Morris 1987, Wiens 1989c, Orians and Wittenberger 1991). As the only way to achieve a good overview of a process is to consider the multiple hierarchical levels of that process (Turner et al. 2001), we intended to test the association between mixed coniferous-deciduous forest cover of eastern Canada (Québec and Ontario) and forest bird occurrence at a broader scale than previous studies, using a grain of 100 km2 and an extent of 155 100 km2. As others studies have suggested (Steele 1992, Böhning-Gaese et al. 1993, Pribil and Picman 1997), broader scales may show different patterns than those found at finer scales, and than those found in other research areas. Our hypothesis was that mixedwood forest possesses is own avifauna and that it particularly attracts certain species of birds, that it is not simply an ecotone regrouping bird species associated to coniferous or deciduous forests.
We used the data from the Atlases of the breeding birds of Quebec (Gauthier and Aubry 1995) and Ontario (Cadman et al. 1987). Both used 10 x 10 km square sampling units based on the UTM map system to inventory the presence of the different species of birds. Quebec’s atlas had surveyed birds for six years (1984-1989) and Ontario’s Atlas for five (1981-1985). These data were mostly collected by volunteers who were instructed to try to cover all habitat types present in sampled squares, for a minimum duration of 16 h (Québec) or 20 h (Ontario). The coverage threshold fixed was based on the asymptote of a non-linear relationship between the number of bird species observed and sampling hours. Coverage thresholds were determined separately for each of the bioclimatic regions in Québec (Gauthier and Aubry 1995) and set to a minimum of 20 h in Ontario (the maximum coverage being of 925 h with an average of 65 h).
Only native forest birds that were present in at least 5% of the selected squares were considered in the analysis. We used Godfrey’s (1986) ‘Habitat’ description to distinguish forest birds from open area birds. A total of 71 species of birds were included.
Québec’s land cover map was produced by the Ministry of Natural Resources of Quebec (Grenier et al. 1994). This map was created from satellite images recorded in 1993 and 1994 from two sources, a part from LANDSAT-5 Thematic Mapper (Landsat-TM) with a resolution of 30 m, and another part from Landsat-MSS with a resolution of 80 m. These images were resampled to produce a standardized mosaic map offering a resolution of 125 m. Twelve land cover classes were distinguished on that map.
Ontario’s land cover map was produced by the Ontario Ministry of Natural Resources (Anonymous 1999). This map was created based on digital, multispectral LANDSAT-5 Thematic Mapper (Landsat-TM) images recorded on a range of dates between 1986 and 1997 with the majority recorded in the early 1990s. The data were sampled to a 25 m pixel size. The maps produced represented 28 land cover classes using a supervised classification method.
We reclassified both Landsat maps to obtain five classes of land cover: 1) coniferous forest, 2) mixedwood forest, 3) deciduous forest, 4) non-forest, 5) open water and unclassified land. Then we calculated the total area per atlas square of each land cover class. Only squares (10 x 10 km) with at least 50% forest cover were considered in the analysis (n=1551) (Figure 1.1). Figure 1.2 shows the frequencies of the different classes of area per forest cover type. Generally, a good range of areas was present, except for the coniferous forest cover and the mixedwood forest cover which had relatively few squares with large surfaces (respectively 51 and 121 squares with more than 70% of cover).
We measured the association between mixedwood, coniferous and deciduous forest areas and bird species occurrence in atlas squares. To account for spatially autocorrelated bird distribution patterns, we modeled the effect of the spatial structure of bird distribution (Borcard et al. 1992, Legendre 1993). We interpolated spatial structure of bird distribution using a second-degree polynomial equation of the x and y coordinates of the atlas square centers, as proposed by Legendre and Fortin (1989) and Borcard et al. (1992) and extracted its associated variance from the data. To do so, we regressed, using the SAS program (SAS Institute Inc 1999-2001) and a binomial distribution, species occurrence within atlas squares in relation with the spatial structure equation and saved the regression residuals obtained. We then used these residuals to do three other regressions (using a normal distribution), each in relation with the proportion of forest found within the atlas squares and with the proportion, within that forest cover, of either mixedwood, coniferous or deciduous covers. To select the cover best predicting species occurrence, we used the information-theoretic approach of Burhnam and Anderson (2002). We calculated the second-order Akaike Information Criterion (AICc) using the model log-likelihood, the number of model parameters (including the intercept), and the sample size ( n ). AICc is an estimator of the strength of evidence of a model. Measuring AICc differences (∆AICc) between models allows the identification of the best model within the selected set, the smallest AICc corresponding to the best model, and permits the ranking of the rest of the models relatively to the “best” model. Based on Burnham and Anderson (2002) recommendations, we interpreted as poor predictors cover types that presented a model ∆AICc superior to 4. We used the ∆AICc and the relation sign obtained for the three models to identify what cover types were favoured at a regional scale.
Because we were studying habitat use by birds at a very broad scale, that our data sources corresponded to different sampling years and to different map resolutions that could influence the definition of land cover classes, we also checked for possible differences in species distribution patterns between data sources. We added to the three existing cover models made from residuals, three new models in which we incorporated a new variable, the interaction factor of the data set sources with each of the cover variables. We calculated the AICc of the new models and compared them to the three other cover models using their ∆AICc. When none of the models including the interaction factor distinguished themselves (∆AICc <4) from those without the interaction factor, we concluded that species distribution patterns in both data sources were not distinct enough to be considered different. Otherwise, we considered species distribution patterns different. For those cases, we plotted, for the two data sources, the regression lines of the relation between the cover type proportion and the residuals of the occurrence-spatial structure relation.
Figure 1.1: Map of the squares of the Atlases of the breeding birds of Quebec and of Ontario. The white squares have sufficient sampling effort. The dark squares have less than 50% of open water area and/or of non-forest area (those used for the analysis).
Figure 1.2: Range of the different forest cover type areas within the 1551 atlas squares analysed. A) Mixedwood forest cover, B) Deciduous forest cover, and C) Coniferous forest cover.
To evaluate to what extent species distribution was associated with land cover, we partitioned the variance present within our data set between our two principal factors of interest (Legendre 1993), i.e., land cover and spatial structure. To do so, we ran five regressions for each bird species, using again the residuals of a first model as the dependent variable for a second model (see above) (Legendre 1993). To measure the variance fraction explained by the factors we used the R-square statistic when using linear regression and Cox and Snell’s Generalized Coefficient of Determination when using logistic regression (SAS Institute Inc 1999-2001). A first regression (using a binomial distribution) with the full model (all forested cover types and the second-degree polynomial equation of the x and y coordinates of the atlas square centroids) allowed us to evaluate the total variance fraction explained by the two factors. Then, to evaluate the variance fraction uniquely associated with the studied land covers we extracted the variance associated with the spatial structure through a first regression (using a binomial distribution) then we used the residual from this regression to do a second regression (using a normal distribution) with the forested covers as the explanatory variables. To evaluate the variance fraction uniquely associated with the spatial structure we inverted variables in the above step. Since a variance fraction was shared by both factors, we substracted from the variance obtained from the full model each variance fraction previously obtained for land cover and spatial structure to calculate that shared fraction. Finally, subtracting the full model variance from 1, gave us the unexplained fraction of the variance.
Mixedwood forest predicted best the occurrence or the absence of 26 bird species through 8 different profiles (Table 1.1). Ten species were positively associated to mixedwood cover because it was the only good predictor (∆AICc >4) of their occurrence. Among them, the most distinctive associations were found for the Evening Grosbeak (see annexe A for the latin names of all studied species), the Pine Warbler, the Red Crossbill and the Magnolia Warbler, based on the high ∆AICc value obtained by the coniferous and deciduous cover types. Fifteen other species were positively linked to mixedwood forest because of the relation sign found but showed more complex profiles since more than one cover was a good predictor of their occurrence (∆AICc <4). Five species showed positive associations with mixedwood and negative relations with both coniferous and deciduous forest covers (e.g. Brown Thrasher, Black-throated Green Warbler, Pileated Woodpecker). Seven species showed a positive association with mixedwood cover but a negative one with deciduous cover (e.g. Dark-eyed Junco, Red-breasted Nuthatch, Brown Creeper), whereas three species presented a positive association with mixedwood but a negative one with coniferous cover (Hairy Woodpecker, Northern Saw-wet Owl, and Ruffed Grouse). Only the Pine Grosbeak occurrence was best predicted (∆AICc <4) by its avoidance of mixedwood forest cover.
Mixedwood forest in combination with other cover types was also positively associated to the occurrence of eight other species. The Hermit Thrush, the Merlin, and the Sharp-shinned Hawk were associated to both mixedwood and deciduous cover types whereas five species occurrences were positively associated with both mixedwood and coniferous covers (e.g. Bay-breasted Warbler, Winter Wren, Cape May Warbler).
The coniferous forest and deciduous forest solely were associated with several species through different profiles. Overall, six bird species were positively associated with coniferous forest and 15 negatively whereas 14 bird species were positively associated with deciduous forest and one, the Gray Jay, negatively. The Tennessee warbler was positively associated to both deciduous and coniferous forests but negatively associated with mixedwood forest.
Of the 71 studied species, 15 showed an occurrence that was well predicted in models that included not only a cover type but also the interaction factor of that cover with the data sources (Table 1.2) meaning that they were presenting distribution pattern differences between the two data sets (Ontario vs. Québec) (Figure 1.3). Among them, the Northern Parula and the Yellow-bellied Flycatcher had a positive relation with mixedwood cover in both data sets, but a stronger relationship in Québec, whereas the Rusty Blackbird showed contradictory relationships with mixedwood cover between data sets (Figure 1.3). The Solitary Vireo, which was classified as a species associated with mixedwood cover in the previous analysis (Table 1.1), showed different responses in Ontario and Québec to the deciduous and the mixedwood covers (Table 1.2). Its negative relation with the previous cover and its positive relation with the latter cover were stronger for the Ontario data set (Figure 1.3).
Table 1.1: Model ranking to measure the strength of cover type in predicting the occurrence of different species of birds within the Atlas squares of Ontario and Quebec, Canada. Second-order Akaike Information Criterion differences (∆AICc) relative to the best model (∆AICc = 0) are shown for each cover type (mixedwood forest (M), coniferous forest (C), deciduous forest (D), the relation sign obtained for strong models (∆AICc value inferior to 4), and the preferred cover type based on ∆AICc value and relation sign.
|
Table 1.2: Influence of data source (Québec vs. Ontario) on species distribution models with cover as predictive variable. Second-order Akaike Information Criterion differences (∆AICc) relative to the best model (∆AICc = 0) are shown for each cover type (mixedwood forest (M), coniferous forest (C), deciduous forest (D). In bold, species for which interaction models perform better (∆AICc <4) than simple cover models for at least one cover type.
|
Globally, the proportion of variance explained by the global model was low. Together, the spatial structure of the bird distribution and the different cover types explained 3 to 56% of the total variance (mean = 21%; Figure 1.4). The variance explained by spatial structure varied from 1% (Sharp-shinned Hawk) to 50% (Great Crested Flycatcher), with a mean of 16% and the variance explained by cover varied from 0 (Yellow-throated Vireo, White-breasted Nuthatch, Whip-poor-will, Red-tailed Hawk, Red-shouldered Hawk, Cooper's Hawk and Alder Flycatcher) to 9% (Evening Grosbeak) with a mean of 2%. Part of the total variance explained was impossible to distribute between those two factors. That variance represented 0 to 17% of the total variance, with a mean of 4%.
Figure 1.3: Occurrence residuals of bird species, after extraction of the data spatial structure variance, in relation with the cover proportion present within the squares of the two atlas, where cover corresponds to mixedwood forest (A), to deciduous forest (B), and to coniferous forest (C), solid lines to Québec data and dotted lines to Ontario data. Species codes refer to Black-and white Warbler (BWWA), Blue Jay (BLJA), Canada Warbler (CAWA), Chesnut-side Warbler (CSWA), Golden-crowned Kinglet (GCKI), Gray Jay (GRJA), Hairy Woodpecker (HAWO), Northern Parula (NPPA), Red-tailed Hawk (RTHA), Rose-breasted Grosbeak (RBGR), Rusty Blackbird (RUBL), Solitary Vireo (SOVI), Spruce Grouse (SPGR), Yellow-bellied Flycatcher (YBFL), and Yellow-bellied sapsucker (YBSA).
Figure 1.4: Variance partitioning (%) of species occurrence among the data spatial structure (pale grey strips), the forested cover type (white strips), the interaction of the spatial structure and the forested cover (black strip), and the unexplained variance (dark grey strips)
More than one third of species studied showed specific associations with mixedwood forest. Some are attracted by it, others avoid it, but undeniably these species are not insensitive to its presence in the region which supports our hypothesis that mixedwood forest possesses its own avifauna. We showed that the proportion of mixedwood cover within a region favoured the occurrence of at least 25 species and limited the one of the Pine Grosbeak. These results support observations made at finer scales. Hobson and Bayne (2000), associated similarly to us Black-throated Green Warbler, Pine Siskin, Red-breasted Nuthatch and Swainson’s Thrush to mixedwood stands. Kirk et al. (1996), using a different classification, suggested that the density of Blackburnian Warbler, Solitary Vireo, Pine Siskin, Yellow-bellied Flycatcher, Pileated Woodpecker, Red-breasted Nuthatch, Ruby-crowned Kinglet and Brown Creeper was more important in mixedwood forest with high coniferous component than in coniferous or in more deciduous sites. These authors also indicated that the density of Evening Grosbeak, Magnolia Warbler, Swainson’s Thrush, Black-throated Green Warbler and Hairy Woodpecker was superior in mixedwood forest with high deciduous component than in the other forest types.
Several species were also sensitive to the amount of coniferous or deciduous forest within their region (100 km2). Others were associated with more than one cover type, including mixedwood forest. These observations therefore support the alternative hypothesis that mixedwood forest may attract certain species of bird because of their partial composition of either coniferous or deciduous trees and that they represent for these species a sub-optimal habitat located at the edge of their optimal habitat, an ecotone.
Only few squares located in the boreal forest presented sufficient observation hours to be included within the analysis. Consequently, the analysis was weakened by the fact that too few squares presenting large areas of coniferous forest cover were included in the sample. This may explain the small number of species found in association to coniferous forest and the high number of species negatively associated with it. A better coverage of coniferous forest could have permitted to classify as species associated to coniferous forest, species that actually present ambiguous status like those associated to both mixedwood and coniferous forests, or species that are negatively associated to deciduous forest according to our current results. Similarly, such improvement could have changed the classification of the species that were actually negatively associated to coniferous forest.
Despite its statistical significance, cover type was not the main driving force behind bird distribution patterns obtained from atlas data. Other factors, ecological as well as methodological, had a greater influence, according to R-square values. Given the coarse nature of land cover and bird data, we did not expect forested cover to explain a large fraction of the total variance. Hobson et al. (2000), attempting to explain the abundance and distribution of 80 bird species within the boreal forest of western Canada, partitioned out the total variance in the species matrix using a canonical correspondence analysis. Similarly to our current results, they only explained small fractions of the variance present in the system: 24% of the variance was explained with non-spatial environmental factors, 3% with spatially structured environmental factors, and 14% with spatial factors, whereas 59% of the variance was unexplained or associated with stochastic fluctuations. In addition to statistical noise, sampling bias could also be present within the analysis since the atlas data used were not corrected for, e.g., access to different cover types or the search for rare bird species.
The present results assess the forest-cover/bird-presence relationship at a regional scale. More specifically, they define the relationship within an extent of more than 155 100 km2, and with a grain adjusted to 10 x 10 km, corresponding to the area of the atlas squares used. Future work should define cover types at different grain sizes to assess grain size influences. It might also be important to check whether the occurrence of the birds is influenced by the spatial pattern of the different cover types. Given that mixedwoods are essentially places where deciduous and coniferous patches meet, true mixedwood birds should respond to the amount of edge between deciduous and coniferous cover types. The latter point is of particular interest also because recent work demonstrated that not only “hard” edges (ex: clear-cut/forest or field/forest) with high contrast could influence birds but also “soft” edges (e.g. an old forest/ a young forest or deciduous forest/ coniferous forest) (Hawrot and Niemi 1996, Fenske-Crawford and Niemy 1997, Penhollow and Stauffer 2000, Fletcher and Koford 2003).
In spite of the coarseness of the data, the large sample sizes allowed us to support the idea that mixedwood forests are more than the sum of their deciduous and coniferous components since some species of birds are particularly influence by the amount of mixte forest in their region. The results obtained are particularly interesting because of the extent of the data set, more than 247 000 hours of field work spread over six years, and because such broad scale analyses remain rare. Atlas data have been previously used to study avian range changes with time and local extinction (Donald and Greenwood 2001), to document the relationship between bird distribution and climate (Venier et al. 1999), to understand inter-species interaction effects on bird distribution (Mikusinski et al. 2001), to document distribution for conservation purposes (Brown et al. 1995, Williams et al. 1996), and in a different way than in the present study, to investigate some bird associations with their environments (Tobalske and Tobalske 1999, Trzcinski et al. 1999). Now that it is well recognized that the process by which birds select their habitat is a spatially-hierarchic one (Wiens 1989a), atlas data will surely be used more in the future, since they are a good complement to fine-scale studies of bird-habitat relationships (Donald and Fuller 1998).
|
|
|
|
| Introduction générale |
|
Chapitre 2 Are mixedwood forests perceived by birds as a distinct forest type? |