A new approach to ecological land classification for the Canadian boreal forest that integrates disturbances
- First Online:
- Cite this article as:
- Grondin, P., Gauthier, S., Borcard, D. et al. Landscape Ecol (2014) 29: 1. doi:10.1007/s10980-013-9961-2
- 572 Views
Traditional approaches to ecological land classification (ELC) can be enhanced by integrating, a priori, data describing disturbances (natural and human), in addition to the usual vegetation, climate, and physical environment data. To develop this new ELC model, we studied an area of about 175,000 km2 in the Abies balsamea–Betula papyrifera and Picea mariana-feathermoss bioclimatic domains of the boreal forest of Québec, in eastern Canada. Forest inventory plots and maps produced by the Ministère des Ressources naturelles du Québec from 1970 to 2000 were used to characterize 606 ecological districts (average area 200 km2) according to three vegetation themes (tree species, forest types, and potential vegetation-successional stages) and four sets of explanatory variables (climate, physical environment, natural and human disturbances). Redundancy, cluster (K-means) and variation partitioning analyses were used to delineate, describe, and compare homogeneous vegetation landscapes. The resulting ELC is hierarchical with three levels of observation. Among the 14 homogeneous landscapes composing the most detailed level, some are dominated by relatively young forests originating from fires dating back to the period centered on 1921. In others, forest stands are older (fires from the period centered on 1851), some are under the influence of insect outbreaks and fires (southern part), while the rest are strongly affected by human activities and Populus tremuloides expansion. For all the study area and for parts of it, partitioning reveals that natural disturbance is the dominant data set explaining spatial variation in vegetation. However, the combination of natural disturbances, climate, physical environment and human disturbances always explains a high proportion of variation. Our approach, called “ecological land classification of homogeneous vegetation landscapes”, is more comprehensive than previous ELCs in that it combines the concepts and goals of both landscape ecology and ecosystem-based management.
KeywordsEcological land classificationEcological gradientsHomogeneous vegetation landscapesNatural disturbancesHuman disturbancesRedundancy analysisK-means clusteringVariation partitioning of vegetation
In natural and human-dominated environments, the interplay between biotic and abiotic forces creates spatial heterogeneity. This heterogeneity generally takes the form of a mosaic composed of patches, each defined by specific relationships between vegetation and environmental conditions (Jenny 1958; Whittaker 1967; White 1979; Urban et al. 1987; Legendre and Fortin 1989; Wu and Loucks 1995; Grossman et al. 1999; Wagner and Fortin 2005). The patches, as represented on a forest map delimiting numerous stands, can be analyzed to define the ecological forces (e.g., latitudinal gradient) structuring the heterogeneity of an area (beta diversity) (Hills 1960; Damman 1979; Zonneveld 1989; Gerardin and Ducruc 1990). Patches can be positioned on an ordination diagram along ecological gradients (continuum concept) and then grouped to form homogeneous vegetation landscapes (community concept) (Whittaker 1967; Ahti et al. 1968; Rowe and Sheard 1981; Zonneveld 1989; Borcard and Legendre 1994; McGarrigal and Cushman 2005). A homogeneous landscape is a portion of land with specific vegetation, physical environment, climate, and disturbance characteristics (both natural and human) (Rowe 1962; Daubenmire 1968; Dufrêne and Legendre 1991). From the macro to the micro scale, a hierarchy of increasingly homogeneous landscapes can be defined (Bailey 1987; Urban et al. 1987; Allen and Hoekstra 1992; Klijn and de Haes 1994; White and Jentsch 2001). At the macro scale, climate is the main driver explaining vegetation distribution (Damman 1979; Bailey 1983; Pojar et al. 1987; Allen and Hoekstra 1992; Payette 1992; Bailey 2009). Bioclimatic zones are at this level (Halliday 1937; Hare 1950; Saucier et al. 2009). At the meso scale, climate, physical features, and disturbances are considered. Finally, at the micro scale, the scale of a field observer, microclimate, physical environment, and disturbances are important variables (Bailey 1987; Lertzman and Fall 1998).
Understanding the causes of spatial heterogeneity and how they vary with scale is a central theme in landscape ecology (Turner 1989; Turner et al. 1993; White et al. 1999). Statistical methods have been developed to test hypotheses about beta diversity (Borcard et al. 1992; Legendre et al. 2005; Peres-Neto et al. 2006; Tuomisto and Ruokolainen 2006). Ecological land classification (ELC), a field related to landscape and numerical ecology, aims to divide large territories into relatively homogeneous landscapes at different levels of observation, and to characterize the contribution of various sets of explanatory variables to vegetation variation. Knowledge of these patterns and processes, and the ecological gradients creating them, is important for ecosystem-based management and biodiversity conservation (Bailey 1983; Gauthier et al. 2001; Legendre et al. 2005; Bailey 2009). Historically, ELC has adopted two main approaches. The first approach is based on vegetation–climate relationships and the second on relationships between vegetation, climate, and the physical environment (Bailey et al. 1985).
The first approach was mainly used from the beginning of the twentieth century to the 1970s. The aim was to identify vegetation changes on mesic soils along ecological gradients (latitudinal, longitudinal, altitudinal) and to associate these changes to macroclimatic variations (temperature and precipitation). Changes in regional (or zonal) vegetation and climate along gradients justified the delineation of bioclimatic units on a subcontinental scale (Halliday 1937; Hustich 1949; Hare 1950; Dansereau 1957; Rey 1960; Küchler 1964; Grandtner 1966; Ahti et al. 1968; Rowe 1972).
The second approach evolved with the influence of geomorphology and a desire to understand ecosystem relationships at a finer scale to help guide resource management decision making. Using ground plots, aerial photographs and satellite images, the physical environment (edaphic-topographic criteria) and its relationship with vegetation and climate were analyzed to delineate biogeoclimatic units. These studies generally covered smaller areas (meso scale) than the vegetation-climate approach (macro scale). From this perspective, within homogeneous ecological units, similar site conditions are expected to support the same type of plant community, including the assemblage of tree species. Together, Betula papyrifera, Populus tremuloides, Abies balsamea, and Picea glauca constitute an example of an assemblage existing on sites with specific combinations of physical features, microclimate, and disturbances. The early-successional stage is dominated by the light-demanding species (e.g., B. papyrifera) of this assemblage, which are progressively replaced in late-successional stages by shade-tolerant species (e.g., A. balsamea), as time since the last disturbance increases. Landscapes with high fire frequencies (e.g., <150 years) are characterized by a large proportion of area occupied by early-successional species (Bergeron et al. 2001; Gauthier et al. 2001; Couillard et al. 2012). This description of forest dynamics corresponds to the notions of potential vegetation (Küchler 1964; Powell 2000; Saucier et al. 2009), habitat type (Daubenmire 1968), site type (Hills 1960; Pojar et al. 1987), and série de végétation (vegetation series) (Rey 1960; Grandtner 1966).
Landscapes can be viewed as a continuum of potential vegetation assemblages along a site gradient (toposequence). Climate–site–vegetation relationships, climatic and edaphic climaxes, site potential and toposequence are key words within this second approach. (Hills 1960; Rowe 1962; Whittaker 1967; Jurdant et al. 1977; Bailey 1983, 2009). The concept of permanence comes into play in this approach, which doesn’t explicitly deal with disturbances.
Although they had received some attention since the beginning of the twentieth century (Clements 1910), natural disturbances have been studied more intensively since 1970 (Heinselman 1973; Payette 1992; Turner et al. 1993; Bergeron et al. 2001). Emphasis has been placed on characterizing the spatial and temporal variability of disturbance regimes (e.g., fire frequency). These studies have demonstrated that forest landscapes are more complex and diversified than originally estimated. Dynamic equilibrium, landscape heterogeneity, and hierarchical patch dynamics are the main concepts structuring these analyses (White 1979; McCune and Allen 1985; Wu and Loucks 1995; Cleland et al. 1997; White et al. 1999; Powell 2000). ELCs were adapted to acknowledge this new approach, which assumes that vegetation is controlled mainly by climate and natural disturbances (Omi et al. 1979; Cissel et al. 1999). From this perspective emerged a landscape classification based on natural disturbance characteristics (fire) in connection with more or less stable vegetation (equilibrium). This approach was mostly applied to larger vegetation landscapes, but could also be considered at a finer scale to the dynamics of a site (Turner et al. 1993; Lertzman and Fall 1998; White et al. 1999).
More recently, human-caused disturbances have been identified as an important driver of landscape dynamics. Anthropogenic activities have altered landscape age class distribution by provoking the loss of old forests (Boucher et al. 2009; Cyr et al. 2009), and have homogenized forest composition by substantially increasing the frequency of early-successional stands on various sites along the toposequence. The relationship between vegetation and the physical environment then becomes much more diffuse than in natural environments (Carleton and MacLellan 1994; Östlund et al. 1997; Lorimer 2001; Schulte et al. 2007; Laquerre et al. 2009). However, ELCs have not yet incorporated human disturbances and are therefore missing an important driver of landscape change.
The first goal of this research is to classify the heterogeneity of the study area within a hierarchy of relatively homogeneous landscapes, on the basis of vegetation and four sets of explanatory variables (climate, physical environment, natural and human disturbances). The second goal is to quantify the proportion of vegetation variation explained by each set and their combinations for some levels of observation in the hierarchical classification.
Sources of information
This study draws on two main contemporary sources of information: forest inventory plots and geospatial databases derived from forest maps. These databases have been developed from Spatial Information on Forest composition based on Tessera (SIFORT). Forest maps from 1970 to 1980 make up the SIFORT-1 database, and forest maps from 1980 to 1990, the SIFORT-2 database. These information sources have been used to characterize 606 ecological districts. The average area of the districts is 200 km2. Each ecological district corresponds to a uniform area described by a specific pattern of surficial deposits, topography, and regional vegetation (Robitaille and Saucier 1998), and was characterized by vegetation (Y-matrix) and explanatory variables (X-matrix).
The Y-matrix was constructed using three complementary vegetation themes, each corresponding to a different aspect or organization level of the vegetation: tree species (e.g., the abundance of Pinus banksiana as a species in an ecological district), forest types (e.g., the abundance of forest stands dominated by Pinus banksiana), and the combination of potential vegetation types and successional stages (e.g., Re2-S2: Picea mariana potential vegetation-Re2 in the early stage of succession-S2) (Appendix 1 in Supplementary Material). The description of tree species and potential vegetation-successional stages is based on forest inventory plots (1970–2000). The SIFORT-2 database was used to describe the forest types.
The X-matrix contains the description of the 606 ecological districts in relation to four sets of explanatory variables (Appendix 2 in Supplementary Material). Climatic variables were estimated using data recorded at meteorological stations and extrapolated to each ecological district by the BioSIM simulator (Régnière 1996). The physical environment was characterized according to the Ministère des Ressources naturelles (MRN) database describing each ecological district, including the relative proportion of area for each surficial deposit type and physiographic variable (Robitaille and Saucier 1998). Natural disturbances were described in terms of the recent history of fires, insect outbreaks, and windthrows over the last 100–150 years. Forest maps (SIFORT-2) were used to evaluate the areas affected by light and severe epidemics, windthrows, and fires. Forest inventory plots (n = 53,635) were used to characterize fire history relative to periods of origin (e.g., 1901–1930). In this article, each of these periods is named according to a year close to its central year (e.g., the 1901–1930 period is referred to as the period centered on 1921). The number of years of infestation by spruce budworm and the frequency of natural fires per 100 km2 from 1938 to 1998 were derived from MRN archives. Human disturbances were described by the relative proportions of agriculture, fallow land, logging, and human-induced fires (from 1938 to 1998). These variables were obtained from forest maps (SIFORT-1 and 2).
Data analysis (Fig. 2)
Constrained analysis, delineation of homogeneous landscapes, and variation partitioning of the vegetation
The ecological land classification of homogeneous vegetation landscapes (ELCH) was developed using a redundancy analysis (RDA) involving a Y-matrix formed by all vegetation themes (tree species, forest types, potential vegetation-successional stages) and constrained by an X-matrix composed of four sets of explanatory variables (climate, natural disturbances, physical environment, and human disturbances). All canonical axes resulting from the RDA were submitted to K-means clustering in order to form groups of ecological districts. K-means clustering allows the formation of two or more groups of ecological districts. While K-means results can be completely hierarchical (smaller groups nested in larger ones), the method does not guarantee this outcome (Legendre and Legendre 2012). This gradual stratification was used to develop a hierarchy of homogeneous landscapes formed of three levels of observation. The landscapes were positioned on an ordination diagram according to the centroid of each landscape, as calculated using the canonical scores of each ecological district. Ordination axis 1 was positioned vertically and axis 2 horizontally, because this configuration represents north–south and east–west gradients in their usual orientation in the study area. The homogeneous landscapes were also characterized using histograms showing the relative proportions of period of origin and disturbance type (fires or insect outbreaks) noted in each forest inventory plot. Variation partitioning of the vegetation was used to quantify the contribution of each set of explanatory variables to vegetation changes along the levels of observation considered in this study (Borcard et al. 1992; Legendre et al. 2005; Peres-Neto et al. 2006; Tuomisto and Ruokolainen 2006). Variation partitioning was performed following the steps proposed by Borcard et al. (2011).
An overview of the vegetation of the study area is used as an introduction to the ELCH. Other descriptions in line with approaches presented in the introduction are described in Appendix 3 in Supplementary Material.
The vegetation of the study area
Ecological land classification of homogeneous vegetation landscapes (ELCH)
RDA and mapping of the scores of the first four canonical axes
The third canonical axis (Fig. 4c) characterizes changes that occur from south-east to north-west. This latitudinal-oblique gradient is strongly linked to changes in the physical environment (Ele variable, Fig. 5c), particularly the transition from a hilly (south-east) to relatively flat topography (north-west). These changes are accompanied by an increase in wetlands (D_7). The fourth canonical axis (Fig. 4d) primarily defines the impact of human activities (Fig. 5d). From approximately 1880 to 1940, land clearing for agricultural settlement had a widespread impact on both the southeastern and southwestern sectors. Beginning in 1905, coal-fired steam locomotives were used in the southern part of the territory to link the agroforestry regions of Abitibi and Lac Saint-Jean (Fig. 1). These activities contributed to numerous human-induced fires (Hf1, Fig. 5, Hardy and Seguin 1984) and promoted changes in both age structure and forest composition. These changes consist mostly in the expansion of P. tremuloides (PotrF) and the presence of many stands originating from the period centered on 1951 (after 1930). During the second half of the twentieth century, mechanized logging spread throughout the A. balsamea–B. papyrifera domain, and gradually towards the north into the Picea mariana–feathermoss domain.
K-means cluster analysis and grouping of ecological districts
Ecological land classification of homogeneous vegetation landscapes (ELCH)
Codes and description of variables used to describe the homogeneous landscapes
Relative area covered by Abies balsamea forest type
Relative basal area for Abies balsamea species
Relative basal area for Acer rubrum species
Relative area covered by Betula papyrifera forest type
Relative area covered by Pinus banksiana forest type
Relative area covered by Picea mariana forest type
Relative area covered by Picea mariana and Abies balsamea forest type
Relative area covered by Populus tremuloides forest type
Relative area covered by Populus tremuloides and Picea mariana forest type
Relative area covered by non-forested wetlands
Annual number of growing degree-days
Rainfall during the growing season (mm)
Absolute difference of topographic elevation (m)
Relative area covered by glaciolacustrine fine-textured (clay) surficial deposits
Relative area covered by organic deposits
Relative area covered by light spruce budworm outbreak (last outbreak 1975–1985)
Relative proportion of forest inventory plots originating from fires between 1901 and 1930
Relative proportion of forest inventory plots originating from fires between 1870 and 1900
Relative proportion of forest inventory plots originating from fires before 1870
Area covered by logging during the 1970s
Frequency of human-induced fires per 100 km2 from 1938 to 1998
Relative area covered by agriculture during the 1970s
Relative area covered by fallow farmland during the 1970s
Relative proportion of forest inventory plots originating from logging since 1930
The transition from 9 to 14 landscapes of level III is mainly justified by disjunctions in geographic distribution (Fig. 7a, b). For example, landscape 12 is divided into three landscapes 121, 122, and 123-pe (pe: Populus expansion), each occupying a specific portion of the study area. Some landscapes are also distinguished in reference to the geographical units delineated on Fig. 6a and b. For example, landscapes 131 and 132 (Fig. 7a), which forms a large unit in Fig. 6c (pale green), is separated into two landscapes in Fig. 6a and b. Landscape 131 is placed high along axis 1 of the ordination diagram (Fig. 7c), revealing its affinities with the Picea mariana-feathermoss domain. Considering the hilly topography, the relative abundance of A. balsamea and its grouping with homogeneous landscape 132 in some analyses (Fig. 3), we classified homogeneous landscape 131 within the northern portion of the A. balsamea–B. papyrifera domain.
Proportion of variation explained by sets of variables along spatial levels of observation (variation partitioning of vegetation)
The total proportion of variation attributed to natural disturbances is relatively high, regardless of geographical entity (Fig. 8, NDt). The unique fractions of variation (e.g., NDu) explained by each set of explanatory variables are generally small, except for natural disturbances. This indicates that some changes in natural disturbances are relatively independent of changes in other sets, especially physical environment. The three large latitudinal bands presented previously (southern, central, northern) and characterizing the natural disturbances are the main examples (Figs. 4b, 6b).
The common fractions of variation explained by the sets of explanatory variables are generally high, especially for natural disturbances. The common variation of natural disturbances is mainly attributed to triple and quadruple combinations. This result confirms that changes in vegetation are closely related to changes in natural disturbances, in combination with other sets. The variation explained by climate (unique and double combinations) is higher for the entire area than the two bioclimatic domains. The impact of human disturbances is generally low, except in the western portion (mainly the Abitibi region), where a high proportion of variation is explained by this set of variables, in combination with the three others (quadruple combination). This indicates that changes in vegetation are closely integrated or dependent on changes in all of the sets, from southern to northern Abitibi (Fig. 3, Appendix 7 in Supplementary Material).
The ELCH supplements ELCs based solely on vegetation
The ELCH considers natural disturbances and other sets of natural explanatory variables
The usefulness of the ELCH lies in its a priori inclusion of natural disturbances and other natural sets of explanatory variables (climate, physical environment). We have shown that natural disturbances are the predominant set explaining vegetation variation (the sum of unique and common variations). The unique variation explained by natural disturbances is considered to be part of the total variation independent of the other sets of explanatory variables. This unique variation might also be defined as the expression of the dominance of natural disturbances over the other sets. This concurs with authors who consider natural disturbances as overlaying other environmental gradients (Heinselman 1973; White 1987; Payette 1992). In addition, this study has shown that natural disturbances vary regionally (White 1987; Mansuy et al. 2010). Some homogeneous landscapes are dominated by young forests, others are older, some are under the influence of insect outbreaks and fires, while the rest are strongly affected by human activities. These landscape types (younger vs. older), based on natural disturbances criteria, are described by specific age class distributions and are in line with those proposed by Turner et al. (1993) and discussed by Lertzman and Fall (1998) and White et al. (1999). Using the vocabulary of these authors, some landscapes of the study area are more stable than others, or are closer to their equilibrium stage.
Although natural disturbances are the main factor, the variation in vegetation explained by this set in combination with others is larger than the unique variation. Consequently, the control of the vegetation by environmental variables is mainly related to the integration of several sets. This paradigm of multiple factors controlling landscape heterogeneity follows numerous authors (Jenny 1958; White 1979, 1987), including those interested by vegetation variation partitioning (Borcard et al. 1992 and following). The integration of multiple factors is maintained for the entire territory and parts of it, all considered at the meso scale (Damman 1979; Bailey 1987; Lertzman and Fall 1998). At this scale, climate is never the dominant set of variables (Ohmann and Spies 1998). We attribute this result to small changes in temperature and precipitation along the latitudinal gradient of the study area. Two bioclimatic domains are present but the same six main species are still present in the landscapes. Climate rather becomes the main driver of landscape heterogeneity in larger territories (macro scale), such as bioclimatic zones (Damman 1979; Bailey 1983; Allen and Hoekstra 1992; Payette 1992; Wu and Loucks 1995; Grondin et al. 2007).
The ELCH considers human disturbances
The ELCH is also novel by its a priori inclusion of human disturbances. We have demonstrated that human activities do not play a major role in explaining the vegetation variation when the entire territory is considered. However, in the southern two ends which are closest to human settlement, four homogeneous landscapes show a strong anthropogenic influence. These four landscapes are well integrated into the sequence of canonical axes formed by the RDA (Fig. 2) and take their place after the sets of natural variables (Figs. 4d, 6c). These results concur with authors who consider human disturbances as an important factor in natural landscapes transformation, and one of the major issues in the context of ecosystem-based management implementation (Urban et al. 1987; White and Mladenoff 1994; Schulte et al. 2007; Boucher et al. 2009). In Abitibi (western portion of the study area), where human activities have had the greatest impact, two homogeneous landscapes (21-pe and 221) are superimposed on a uniform area with respect to the physical environment (mesic clay deposit), climate, and natural disturbances (abundance of fires centered on the period 1921). This indicates that anthropogenic disturbances can generate specific homogeneous landscapes. On mesic clay deposits, the expansion of P. tremuloides and A. balsamea could eventually continue northward, under the influence of logging, to the northern natural limit of P. tremuloides. The abundance of these two species and the consistent decrease of Picea mariana in mixed stands of the Picea mariana-feathermoss domain, favoured by intensive management practices and, possibly, climatic change, could contribute to the northward expansion of the A. balsamea–B. papyrifera domain into the northern boreal forest currently dominated by Picea mariana (Grondin and Cimon 2003; Laquerre et al. 2009; Arbour and Bergeron 2011).
This study builds on the research of authors who have described ecological gradients and used them to define an ELC. Our approach, the ELCH, is original in including, a priori, landscape disturbance patterns (natural and human) as sets of variables. A sequence of numerical analyses (RDA, K-means clustering, variation partitioning) has been used to describe the ecological gradients and define an ELC. We have shown that it is possible to elaborate on an ELC by integrating the main factors structuring landscape heterogeneity, even in areas with great variability of natural disturbances and a strong local influence of human disturbances. Landscape spatial heterogeneity of the study area, considered at the meso scale, is mainly explained by natural disturbances in synchronicity and overlapping with changes in the physical environment, climate, and human activities. Our integrative and quantitative approach of ecological gradients enhance and perhaps slightly modifies our perception and understanding of factors causing landscape heterogeneity in the circumboreal forest zone. This study could not have been carried out without the large databases available at the MRN. More detailed data, especially with respect to natural disturbances (fire origin maps), could lead to slightly different and more accurate results. Other numerical analyses could also be tried (e.g., fuzzy clustering). This first ELCH should be considered as a point of reference to define and compare natural and managed landscapes, to initiate more detailed studies on forest dynamics (e.g., natural variability of homogeneous landscapes), and to estimate the effects of climate changes on vegetation.
Data (plots, maps, archives) used in this study were collected by the staff of the Ministère des Ressources naturelles du Québec (MRN) between 1970 and 2000. This study could not have been produced without the effort of many individuals. Comments by Yan Boucher, David T. Cleland, Paul Jasinski, Jason Laflamme, Del Meidinger, Germain Mercier, 2 anonymous reviewers, as well as stylistic revisions by Debra Christiansen-Stowe, Karen Grislis, and Denise Tousignant were all greatly appreciated. We would also like to thank Denis Hotte and Véronique Poirier for their assistance in data analysis and geomatics. This study was funded by the MRN.