Plusieurs approches tentent de mesurer l'impact des perturbations anthropiques sur les écosystèmes. L'approche BACI (Before-After Control-Impact) consiste à suivre deux groupes de sites (contrôle et impact), avant et après une perturbation, afin de mesurer l'effet de cette dernière sur les écosystèmes. Les études BACI permettent de contrôler la variabilité naturelle entre les groupes de sites, par le suivi des mêmes sites d'impact avant et après la perturbation, tout en minimisant la variabilité naturelle entre les années grâce au suivi de sites de contrôle échantillonnés également avant et après la perturbation. Puisque la variation naturelle entre les années dans les sites d'impact est estimée à partir de celle des sites de contrôle, il est nécessaire de sélectionner des sites de contrôle dont les caractéristiques limnologiques sont semblables à celles des sites d'impact. Ceci est essentiel pour une bonne application de l'approche BACI, afin de s'assurer que les sites naturels et perturbés répondent de la même façon aux variations naturelles interannuelles dans l'environnement et que les différences observées dans les sites d'impact avant et après la perturbation soient attribuables à celle-ci.Cet article propose une méthode de sélection des sites de contrôle dans le cadre d'une étude BACI portant sur l'impact des coupes forestières sur le zooplancton des lacs de la forêt boréale au Québec. Le zooplancton de 16 lacs de la forêt boréale a été échantillonné un an avant (2000) et deux ans après (2001-2002) des coupes forestières sur le bassin versant de certains lacs. Six lacs ont subi des coupes importantes sur 44 à 77 % du bassin versant (lacs de coupe: DA2, DF2, DF7, DF9, K4, K8) et 10 lacs sont restés à l'état naturel ou n'ont subi que des coupes négligeables sur moins de 2 % du bassin versant (lacs de référence: K2, AB34, AB35, AB40, CSL5, DA4, DF4, N35, N89, N43). Parmi ces dix lacs de référence, nous avons sélectionné les six lacs les plus semblables aux lacs de coupe, à l'aide d'analyses en composantes principales (ACP) basées sur la similarité des variables morphométriques, de la qualité de l'eau et du zooplancton avant la coupe (2000). De plus, les variables ayant la plus grande contribution à la variation totale au niveau de ces trois groupes de variables ont été déterminées. Quatre lacs ont été exclus (K2, N89, AB35, AB40) et six lakes (AB34, CSL5, DA4, DF4, N35, N43) ont été sélectionnés comme référence. Finalement, la validité du choix des six lacs de référence a été testée par des analyses de redondance (RDA) avec une variable binaire qui permettait de distinguer les lacs de coupe des lacs de référence sélectionnés. Les analyses de redondance ont montré que les variables de la morphométrie, de la qualité de l'eau et du zooplancton des lacs de référence sélectionnés ne différaient pas significativement de celles des lacs de coupe avant la perturbation. En conséquence, les différences observées après la perturbation dans les six lacs de coupe, relativement aux variations naturelles dans les six lacs de référence sélectionnés, devraient être attribuables à l'effet de la coupe forestière. La méthode de sélection développée dans le cadre de cette étude peut être utilisée pour évaluer à l'aide d'une approche BACI les effets de toute perturbation anthropique sur les écosystèmes
Several designs can be used to assess the effects of human perturbations on ecosystems. However, the main difficulty is to isolate natural sources of variation from the variation induced by the perturbation. Several studies have shown that the natural differences between the reference and the impacted sites may influence their responses to the perturbation. In a comparative design comparing the conditions at the reference sites and the impacted sites after the perturbation, it is not possible to control for the natural sources of variation between these two groups of sites that occurred before the perturbation. This natural variation among sites is taken into account in a Before-After design in which the same sites are monitored before and after perturbation, but the natural sources of variation among years cannot be separated from the variation induced by the perturbation. In this study, we used a BACI (Before-After Control-Impact) design to measure the effect of a perturbation on an ecosystem by following two groups of sites (control and impacted) before and after the perturbation. A BACI design is the only one that allows controlling for the natural variability among sites by following the same impacted sites before and after the perturbation. This approach also takes into account the natural variability among years by monitoring concomitantly the control sites before and after the perturbation. Since the natural variability among years at the impacted sites is estimated by the variability measured at the control sites, it is essential to select control sites that are the most similar to the impacted sites with respect to their limnological characteristics. This is a requirement for the good use of the BACI design, to make sure that control and impacted sites have the same responses towards year-to-year natural variations in their environment and to ensure that the differences observed at the impacted sites before-after the perturbation are caused by it.This paper proposes a method for the selection of control sites in a BACI design to study the impact of forest harvesting on zooplankton communities in lakes of the boreal forest in Québec. Zooplankton in 16 lakes was sampled one year before (2000) and two years after (2001-2002) forest harvesting in the watersheds of some lakes. Six lakes were impacted by important harvesting on 44 to 77% of their watershed (cut lakes: DA2, DF2, DF7, DF9, K4, K8) and ten lakes were considered as natural lakes with forest harvesting on less than 2% of their watershed (reference lakes: K2, AB34, AB35, AB40, CSL5, DA4, DF4, N35, N89, N43). Among these ten natural lakes, we selected the six reference lakes that had limnological features most similar to the cut lakes, based on the morphometry, water quality and zooplankton variables before forest harvesting (2000). We used principal component analyses (PCA) to compare the lakes (10 natural lakes and six lakes to be impacted by watershed harvesting) using ordination biplots. A PCA was done for each group of variables. Lake volume, maximal depth, and area and the slope of the watershed were the variables having the most important contributions to the total variation in morphometry. Positive correlations were found between the slope of the watershed and the maximal depth of the lake and between the lake area and the lake volume. This PCA allowed us to eliminate two reference lakes that had higher values in lake volume and area for one lake (K2) and in lake volume and maximal depth for the other one (AB35), compared to the cut lakes. When comparing water quality variables, total phosphorus (TP), total nitrogen (TN), dissolved organic carbon (DOC), Secchi depth, Ca2+, Mg2+, alkalinity (ALK) and pH contributed more to the total variation than other variables. Among these variables, DOC, TP and TN were positively correlated together whereas they were negatively correlated to water transparency (Secchi). Two reference lakes (AB35, N89) had higher transparency and lower values in DOC, TP and TN and one reference lake (AB40) had lower values in pH, ALK, Ca2+ and Mg2+ compared to cut lakes. Since one of these three lakes had previously been eliminated on the basis of its morphometric variables, we had a total of four reference lakes excluded from the study. The PCA on zooplankton variables revealed that only the largest class of zooplankton (>500 µm) contributed significantly to zooplankton total variation. In general, reference lakes and cut lakes had a similar distribution on the ordination biplot so no lake was eliminated with respect to zooplankton variables. Finally, four reference lakes were excluded (K2, N89, AB35, AB40) and six reference lakes were selected (AB34, CSL5, DA4, DF4, N35, N43).The validation of the selection of the six reference lakes was made using redundancy analyses (RDA) with a binary variable discriminating between the six cut lakes and the six reference lakes selected in the PCAs. Multivariate methods such as RDA have the advantage of testing if the two groups of lakes are different with respect to many variables and not just one, as would be the case when using a univariate method. Redundancy analyses showed that the morphometric, water quality and zooplankton variables in the six selected reference lakes were not significantly different from those in the six cut lakes before harvesting. The RDA results supported the selection of the six reference lakes based on the ordination biplots from the three PCA made on the three groups of variables. We could then presume that any difference observed after the perturbation in the cut lakes, relative to the natural variation among years in the six reference lakes, could be attributed to the effect of forest harvesting. Finally, this method of selection of control sites could be used for any BACI study testing the effects of human perturbation on ecosystems.
