Abstract

Aims

We compare performance of ecosystem classification maps and provincial forest inventory data derived from air photography in reflecting ground beetle (Coleoptera: Carabidae) biodiversity patterns that are related to the forest canopy mosaic. Our biodiversity surrogacy model based on remotely sensed tree canopy cover is validated against field-collected ground data.

Methods

We used a systematic sampling grid of 198 sites, covering 84 km2 of boreal mixedwood forest in northwestern Alberta, Canada. For every site, we determined tree basal area, characterized the ground beetle assemblage and obtained corresponding provincial forest inventory and ecosystem classification information. We used variation partitioning, ordination and misclassification matrices to compare beetle biodiversity patterns explained by alternative databases and to determine model biases originating from air photo-interpretation.

Important Findings

Ecosystem classification data performed better than canopy cover derived from forest inventory maps in describing ground beetle biodiversity patterns. The biodiversity surrogacy models based on provincial forest inventory maps and field survey generally detected similar patterns but inaccuracies in air photo-interpretation of relative canopy cover led to differences between the two models. Compared to field survey data, air photo-interpretation tended to confuse two Picea species and two Populus species present and homogenize stand mixtures. This generated divergence in models of ecological association used to predict the relationship between ground beetle assemblages and tree canopy cover. Combination of relative canopy cover from provincial inventory with other geo-referenced land variables to produce the ecosystem classification maps improved biodiversity predictive power. The association observed between uncommon surrogates and uncommon ground beetle species emphasizes the benefits of detecting these surrogates as a part of landscape management. In order to complement conservation efforts established in protected areas, accurate, high resolution, wide ranging and spatially explicit knowledge of landscapes under management is primordial in order to apply effective biodiversity conservation strategies at the stand level as required in the extensively harvested portion of the boreal forest. In development of these strategies, an in-depth understanding of vegetation is key.

INTRODUCTION

Use of biodiversity surrogates provides a practical, realistic and cost effective basis for landscape scale conservation strategies (Duro et al. 2007; Nagendra 2001; Turner et al. 2003). It offers a robust scientific basis to relate patterns of biodiversity to variation in managed ecosystems (Spence and Langor 2006; Spence et al. 2008) and does so in terms directly relevant to landscape planners. Coupled with remote sensing and geographical information systems, development of biodiversity surrogates can provide the high-resolution maps needed to assist regional conservation planning and monitoring efforts (Ferrier 2002; Margules and Pressey 2000).

Given limited technical and financial resources for identifying and mapping the entire range of biodiversity over large areas, effective surrogates are essential for embracing a holistic ecosystem approach for maintenance of global and local biodiversity (Franklin 1993; Noss 1990; Wilson 1988). Of course, relationships between the biodiversity and the surrogates employed must be monitored through detailed spot checks to ensure their fidelity over time (Spence et al. 2008). Nonetheless, because financial and technical support invested in large-scale bioinventories will always be insufficient to generate biodiversity maps with the grain size, extent and species diversity level required for management planning, development of effective surrogates is our only hope. In this paper, we show how vegetation and ecosystem classification might be employed to achieve this goal for the boreal mixedwood.

An array of remotely sensed data about landscapes on earth is already available and widely used in land management (Richards and Jia 2006). There is compelling scientific evidence for mammal and bird species that remotely sensed environmental variables correlate with observed biodiversity patterns (Leyequien et al. 2007; Rodrigez et al. 2007). Furthermore, evidence for such correlations is now emerging for more cryptic groups such as invertebrates (Barbaro et al. 2007; Eyre and Luff 2004; Kerr et al. 2001; Müller and Brandl 2009), mosses and fungi (McMullan-Fisher et al. 2009). Research to date has concentrated on the detection, description and predictive modeling of relationships between biodiversity parameters and surrogates but has largely ignored assessments of accuracy (Czaplewski 2003; Morgan et al. 2010). Failure to consider such error may have cascading effects on the ecological associations modeled, the consequent conservation strategies implemented and, in the final analysis, the efficacy of natural resource management plans (Thompson et al. 2007).

Aerial photography is especially well suited for ecological land management (Hall 2003; Morgan et al. 2010). For example, in Canada, forest inventories are largely based on interpretation of aerial photographs ranging in scale from 1:10 000 to 1:20 000 (Leckie and Gillis 1995). These interpretations are the backbone of most forest management decisions and are used to inform the basal logic of many scientific predictive models (Avery and Berlin 2003; Paine and Kiser 2003). Stand level data interpreted from such photography are used to refine the spatial resolution of ecosystem classification frameworks (e.g. Beckingham et al. 1999; Nadeau et al. 2004) developed for national and international biodiversity conservation reporting (Bailey and Hogg 1986; Marshall et al. 1996; Olson et al. 2001). In addition to providing guidance for the establishment of biodiversity reserves on coarser scales (Margules and Pressey 2000), ecosystem maps with finer resolution have the potential to guide the detailed development of conservation strategies in operationally managed landscapes. The scale and the detail of the information contained in these maps provide the knowledge base used for stand level management and are therefore a crucial tool for linking these activities to biodiversity patterns across landscapes.

Most studies linking remote sensing to biodiversity generally focus on landscapes containing contrasting habitats such as farm lands, urban area and forest (Barbaro et al. 2007; Eyre and Luff 2004; Kerr et al. 2001; McMullan-Fisher et al. 2009; Müller and Brandl 2009). Thus, by focusing on biodiversity patterns related to canopy heterogeneity, we assess the ability of remote sensing to detect biodiversity patterns in a rather homogeneous forest environment, as is directly relevant to addressing concerns about boreal biodiversity.

In this study, we compare performance of ecosystem classification maps (ecosite; Nielsen et al. 1999) and Alberta Vegetation Inventory (AVI) data (http://www.srd.alberta.ca/MapsPhotosPublications/Maps/ResourceDataProductCatalogue/ForestVegetationInventories.aspx) in reflecting biodiversity patterns for ground beetles (Coleoptera: Carabidae). We first assess how much variation in the beetle community can be explained by each of these data sets and compare these results with data obtained through a detailed ground survey of trees. We then point out dissimilarities in the ecological models originating from remotely acquired data and ground surveys. Finally, we assess the accuracy of data originating from air photos by directly comparing the AVI data set with the ground survey data collected on a particular landscape for each tree species.

MATERIALS AND METHODS

Study site

The study was conducted at the Ecosystem Management Emulating Natural Disturbance (EMEND; see Spence et al. 1999) research site located in the lower foothills ecoregion of the mixedwood boreal forest (Beckingham et al. 1996) in northwestern Alberta. The dominant tree species are Picea glauca (Moench) (white spruce), Populus tremuloides Michx. (aspen) and Populus balsamifera L. (balsam poplar) on well drained sites and Picea mariana (Mill.) (black spruce) and Larix laricina (Du Roi) K. Kock (tamarack) on poorly drained sites. The elevation varies between 677 and 880 m in a catena-like topography of rolling hills, mostly consisting of morainal deposits with extensive valleys and depressions covered by lacustrine and organic deposits (Alberta Environmental Protection 1994).

Methods

In the summer of 2002, we established a systematic grid of 200 sampling sites covering 84 km2 of mixedwood forest. Sites were located roughly 640 m apart, with grid points adjusted locally within the nearest stand to include trees over 5 cm of diameter at breast height (dbh) and to be at least 40 m from any anthropogenic disturbance. In each site, we recorded the species (according to Moss 1983) and dbh for the 25 living trees, >5 cm dbh, closest to the site center. dbh and species were also recorded for one stem of any additional tree species detected within 50 m of the center. This allowed us to include tree species absent from the 25 sampled stems that also contribute in explaining the ground beetle biodiversity at the sampled site. Two sites were omitted from the grid because large harvested areas would have placed these sampling sites within 40 m to the next closest site.

Restricting the site selection to forest with well-developed trees (>5 cm dbh) allowed us to focus on the surrogate variable ‘relative tree canopy cover’, as derived from aerial photography. It also allowed us to check the ability of remote sensing to depict biodiversity patterns that might be related to stand level boreal forest canopy heterogeneity.

During summer 2003, we used three pitfall traps at each site to sample the ground beetle (Coleoptera: Carabidae) assemblage. A trap consisted of a plastic cup with an opening diameter of 11 cm and a depth of 13 cm, containing a plastic inner cup and a wooden roof supported over the trap by two nails (Spence and Niemelä 1994). Traps were installed 15 m from the center of the site at bearing of 0, 120 and 240°. Silicate-free ethylene glycol (GM Dex-Cool®) in the inner cup was used as a killing agent and preservative. Traps were open from the second week of May until the third week of August for a maximum of 99 potential sampling days embracing most of the frost-free period. Trap contents were collected four times over this period. Five sites established in 2002 were harvested during the following winter and therefore were omitted from ground beetle sampling. All carabid specimens were identified to the species level according to Lindroth (1961, 1963, 1966, 1968, 1969), with nomenclature following Bousquet (1991). Voucher specimens are deposited in the Spence laboratory collection and the Strickland Entomological Museum of the University of Alberta.

Data about relative tree canopy cover were obtained from the province wide AVI Phase 3 for each site, using ArcView 3.4 software (ESRI®). These data consist of photo-interpretations of relative canopy cover in 10% classes for each tree species recorded in the delineated stand (polygon). Tree data collected on the ground were transformed into relative basal area and grouped into 10 categories of 10% plus a 0% category in order to be comparable with the relative tree canopy cover based on AVI data. Continuous data about relative basal area included within plus or minus 5% of the category label were merged into this category to approximate the categorization of relative canopy cover made by air photo-interpreters in developing the AVI data. The 0% category therefore includes values of relative basal area comprised between 0 and 5%, the 10% category includes values between 6 and 15% and so on. We assumed that relative basal area of tree species is proportional to the relative canopy cover because of the linear correlation between diameter at breast height and crown diameter (Spurr 1960).

The ecosite classification data were acquired from a geo-referenced digital map generated by Nielsen et al. (1999) using the SiteLogix ecological mapping system (Beckingham et al. 1999) and following the ecological land classification system for west-central Alberta (Beckingham et al. 1996). This classification system includes four hierarchical levels: natural subregion (defined by climatic variation of reference physiographic sites), ecosite (defined by a nutrient-moisture gradient within subregions), ecosite phase (defined according to the main canopy cover within ecosites) and plant community (defined according to the understory plants within ecosite phases). We used the ecosite phase level because it provided the highest resolution available through the mapping procedure (Nielsen et al. 1999). To obtain ecosite phase variables for each site, we calculated the area in square meters for each ecosite phase within a 50 m radius of the center of the sampling site using ArcView 3.4 (ESRI®).

Statistical analysis

Beetle samples from the three traps in each site were pooled and divided by the total number of effective trapping days to standardize for sampling effort. Samples from traps not operating (disturbed by animals or flooded) at time of sample collection were excluded from further analysis.

After a Hellinger transformation of the beetle species data (Legendre and Gallagher 2001), a variation partitioning procedure (Borcard et al. 1992; Legendre and Legendre 1998) was used to determine the proportion of variance in ground beetle assemblages explained by AVI, ecosite map and the ground survey data. These analyses were performed using the vegan (Oksanen et al. 2011) package within the R statistical language (R Development Core Team 2011) on the 193 sites from which beetle data were available.

Association between land cover variables from each survey method and the beetle assemblage was assessed by performing a canonical redundancy analysis (RDA, Legendre and Legendre 1998) on the transformed beetle data using the R package rdaTest (Legendre and Durand 2010). The Hellinger transformation minimizes the inward folding of the environmental gradient extremes (Legendre and Gallagher 2001), reducing the typically strong ‘horseshoe effect’ in such data (Legendre and Legendre 1998).

Ordinations for each survey method were projected in the beetle assemblage (used as the response variable) space (Legendre and Legendre 1998) to facilitate comparison of the ability of alternative land survey data to explain biodiversity patterns. Vectors for the 13 most abundant beetle species and all tree species were added to help interpret drivers of the ordination results. Ordinal axes were judged as significant (P < 0.05) predictors of beetle assemblages based on 999 random permutations. Forward selection (Blanchet et al. 2008) was performed on the ecosite data set in order to choose the variables explaining the most variation in the beetle assemblage, using the R package packfor (Dray et al. 2009).

Provincial AVI and ground data for the four most abundant tree species were compared for 198 sites, using misclassification matrices (Congalton 2001; Foody 2002; Stehman and Czaplewski 1998; Wulder et al. 2006). We used all sites where tree data were available in order to increase the power of this analysis, instead of restricting our analysis to the 193 sites where beetle data were also available. AVI data classified within two categories (±20%) from the ground data were considered accurate. This range was determined based on the fact that many factors such as sampling design, positional error from our Global Positioning System unit, the loose linear relationship between basal area and crown diameter and the methodological procedure of photo-interpretation may all induce variability in the results. The average relative proportion of each tree species in misclassified sites was calculated for AVI and ground data in order to determine tree species biases in AVI. A t-test, performed with 999 permutations, was carried out for each tree species in misclassified sites to calculate significance of divergence in the relative proportion estimates between the two inventory methods.

RESULTS

Overall, 9776 ground beetles identified as representing 41 species were collected across the 193 sites. Although 5039 individual trees were sampled over the 198 sites, trees were represented by just eight overstory species: white spruce, aspen, black spruce, balsam poplar, balsam fir (Abies balsamea (L.)), tamarack, paper birch (Betula papyrifera Marsh.) and lodgepole pine (Pinus contorta Dougl. ex Loud). These same tree species were recorded on this landbase in the AVI. Twenty-three ecosite classification categories were detected in the study, all within a 50 m radius of at least one site where beetles were sampled (Table 1). Some records must be attributed to a combination of two ecosite phases originally described by Beckingham et al. (1996) because the mapping process could not unequivocally attribute a dominant ecosite class to the polygon (Nielsen et al. 1999).

Table 1:

code, total area, proportional area and number of sites for every land classification category recorded within a 50 m radius of our 193 sites

Ecosite phase Code Total area (m2n sites 
Low-bush cranberry Sw e4a 423 130 27.9 77 
Low-bush cranberry Aw e2a 249 251 16.4 46 
Low-bush cranberry Aw-Sw-Pl e3a 241 713 15.9 59 
Treed bog k1a 145 549 9.6 35 
Treed poor fen//treed bog l1/k1a 91 588 6.0 19 
Low-bush cranberry Sw//bracted honeysuckle Sw e4/f4 84 896 5.6 16 
Shrubby poor fen//shrubby meadow l2/g1a 78 144 5.2 21 
Labrador tea/horsetail Sb-Sw//labrador tea-mesic Pl-Sb j1/d1a 27 740 1.8 
Low-bush cranberry Aw//bracted honeysuckle Aw-Pb e2/f2a 26 874 1.8 
Labrador tea/horsetail Sb-Sw//horsetail Sw j1/i3 20 278 1.3 
Treed bog//labrador tea-subhygric Sb-Pl k1/h1a 19 044 1.3 
Labrador tea-mesic Pl-Sb//low-bush cranberry Pl d1/e1a 18 358 1.2 
Bracted honeysuckle Aw-Pb f2a 18 358 1.2 
Low-bush cranberry (no tree canopy) e?a 14 178 0.9 
Labrador tea-subhygric Sb-Pl//labrador tea/horsetail Sb-Sw h1j1a 9790 0.6 
Water w1 9749 0.6 
Low-bush cranberry Sw//labrador tea/horsetail Sb-Sw e4j1 9152 0.6 
Treed poor fen l1a 7903 0.5 
Bracted honeysuckle Sw//horsetail Sw f4i3 6119 0.4 
Anthropogenic z1 5640 0.4 
Low-bush cranberry Pl e1 5002 0.3 
Bracted honeysuckle willow f6a 1857 0.1 
Shrubby poor fen//forb meadow l2g2 1635 0.1 
Ecosite phase Code Total area (m2n sites 
Low-bush cranberry Sw e4a 423 130 27.9 77 
Low-bush cranberry Aw e2a 249 251 16.4 46 
Low-bush cranberry Aw-Sw-Pl e3a 241 713 15.9 59 
Treed bog k1a 145 549 9.6 35 
Treed poor fen//treed bog l1/k1a 91 588 6.0 19 
Low-bush cranberry Sw//bracted honeysuckle Sw e4/f4 84 896 5.6 16 
Shrubby poor fen//shrubby meadow l2/g1a 78 144 5.2 21 
Labrador tea/horsetail Sb-Sw//labrador tea-mesic Pl-Sb j1/d1a 27 740 1.8 
Low-bush cranberry Aw//bracted honeysuckle Aw-Pb e2/f2a 26 874 1.8 
Labrador tea/horsetail Sb-Sw//horsetail Sw j1/i3 20 278 1.3 
Treed bog//labrador tea-subhygric Sb-Pl k1/h1a 19 044 1.3 
Labrador tea-mesic Pl-Sb//low-bush cranberry Pl d1/e1a 18 358 1.2 
Bracted honeysuckle Aw-Pb f2a 18 358 1.2 
Low-bush cranberry (no tree canopy) e?a 14 178 0.9 
Labrador tea-subhygric Sb-Pl//labrador tea/horsetail Sb-Sw h1j1a 9790 0.6 
Water w1 9749 0.6 
Low-bush cranberry Sw//labrador tea/horsetail Sb-Sw e4j1 9152 0.6 
Treed poor fen l1a 7903 0.5 
Bracted honeysuckle Sw//horsetail Sw f4i3 6119 0.4 
Anthropogenic z1 5640 0.4 
Low-bush cranberry Pl e1 5002 0.3 
Bracted honeysuckle willow f6a 1857 0.1 
Shrubby poor fen//forb meadow l2g2 1635 0.1 
a

Ecosite phases selected by foward selection and presented in Fig. 2.

The names and codes follow nomenclature by Nielsen et al. (1999).

Relationships between beetles and forest composition

Ecosite phase explained higher variance in the carabid assemblage (34%) compared to ground survey (32%) and AVI (22%, Fig. 1). The majority of variance in the beetle assemblage explained by the AVI data (18%) was congruent with the ground survey data and was entirely included in the variance explained by the ecosite classification system. The ecosite phase and ground survey data sets mutually explained 23% of the variance in the beetle assemblage with ecosite phase explaining an additional 11% on its own.

Figure 1:

Venn diagram representing the proportion of the variation in the beetle community explained by the different data sets. Percentages are rounded to the closest unit.

Figure 1:

Venn diagram representing the proportion of the variation in the beetle community explained by the different data sets. Percentages are rounded to the closest unit.

The ecosite map does reflect ground beetle assemblage in a way that a specific pattern of association between ecosite phases and beetle species is discernable (Fig. 2). However, sites on the positive side of the first axis are dominated by ecosite phase characteristic of mesic sites, while the sites on the negative side of the first axis are dominated by ecosite phase characteristic of wetter areas (Beckingham et al. 1996). The deciduous dominated ecosites phases (e2, e2f2, f2, e? and f6) are grouped in the third quadrant together with those for the beetles Platynus decentis (Say), Agonum retractum LeConte, Trechus chalybeus Dejean and Patrobus foveocollis (Eschscholtz). The vector for the mixed ecosite phase (e3) is oriented along the first axis at the edge of quadrant one and is closely aligned to the vectors for Pterostichus adstrictus Eschscholtz and Calathus ingratus Dejean, two of the most common carabids on this landscape. The vector for the white spruce dominated ecosite phase (e4) is in the first quadrant together with those for Calathus advena (LeConte), Stereocerus haematopus (Dejean) and Pterostichus brevicornis (Kirby). Pterostichus punctatissimus (Randall) is associated with ecosite phases having a poor nutrient regime and, despite its often stated affinity for wet ecosite phases (k1, l1k1, k1h1), the analysis shows that this beetle also occurs in ecosite phases typical of drier moisture regime (d1/e1, j1d1, Beckingham et al. 1996). Agonum gratiosum Mannerheim and Platynus mannerheimii Dejean are also associated with wet ecosites phases but these sites are generally richer in nutrient availability as reflected by the plant community composition (l2g1, l1).

Figure 2:

ordination diagram resulting from a redundancy analysis between the beetle assemblage and the ecological classification map. Gray dots are the 193 sites. The ecological categories are in light gray and the beetles species are in black. Refer to Table 1 for the code of the ecological categories. The abbreviations of the beetle species are as follow: Agograt: Agonum gratiosum, Agoretr: Agonum retractum, Caladve: Calathus advena, Calingr: Calathus ingratus, Carcham: Carabus chamissonis, Patfove: Patrobus foveocollis, Pladece: Platynus decentis, Plamann: Platynus mannerheimii, Pteads: Pterostichus adstrictus, Ptebrev: Pterostichus brevicornis, Ptepunc: Pterostichus punctatissimus, Stehaem: Stereocerus haematopus, Trechal: Trechus chalybeus.

Figure 2:

ordination diagram resulting from a redundancy analysis between the beetle assemblage and the ecological classification map. Gray dots are the 193 sites. The ecological categories are in light gray and the beetles species are in black. Refer to Table 1 for the code of the ecological categories. The abbreviations of the beetle species are as follow: Agograt: Agonum gratiosum, Agoretr: Agonum retractum, Caladve: Calathus advena, Calingr: Calathus ingratus, Carcham: Carabus chamissonis, Patfove: Patrobus foveocollis, Pladece: Platynus decentis, Plamann: Platynus mannerheimii, Pteads: Pterostichus adstrictus, Ptebrev: Pterostichus brevicornis, Ptepunc: Pterostichus punctatissimus, Stehaem: Stereocerus haematopus, Trechal: Trechus chalybeus.

The tree species vectors projected on the ordination for the AVI (Fig. 3) provide groupings that are generally similar to those in the ordination for the ground survey data (Fig. 4). For both ordination, (i) the deciduous tree species group together in the fourth quadrant, (ii) white spruce, lodgepole pine and balsam fir (A. balsamea) are found on the positive side of the second axis and (iii) black spruce and tamarack group in the third quadrant. However, the relative positions of some tree species differ between these data sets. For example, the ground survey data achieve a greater differentiation between the aspen and the balsam poplar vectors (cf. Figs 4 and 5) than in the ordination based on the AVI data (Fig. 3). Similarly, differentiation between tamarack and black spruce is less pronounced in the ordination based on the provincial inventory data. We also noted that the angles between the white spruce vector and those for both aspen and balsam poplar are more acute for the ground survey data than for the AVI data. Furthermore, the relative position of vectors for balsam fir and lodgepole pine differs between data for AVI and ground surveys.

Figure 3:

ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative canopy cover of provincial forest inventory map. The abbreviations for beetle species are as in Fig. 2. The abbreviations for tree species are as follow: Aw: Populus tremuloides, Bw: Betula papyrifera, Fb: Abies balsamea, Lx: Larix laricina, Pb: Populus balsamifera, Pl: Pinus contortae, Sb: Picea mariana, Sw: Picea glauca.

Figure 3:

ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative canopy cover of provincial forest inventory map. The abbreviations for beetle species are as in Fig. 2. The abbreviations for tree species are as follow: Aw: Populus tremuloides, Bw: Betula papyrifera, Fb: Abies balsamea, Lx: Larix laricina, Pb: Populus balsamifera, Pl: Pinus contortae, Sb: Picea mariana, Sw: Picea glauca.

Figure 4:

the first and the second axis of the ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative basal area of tree species recorded from ground survey. Abbreviations for the beetle species are as in Fig. 2 and those for the tree species are as in Fig. 3.

Figure 4:

the first and the second axis of the ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative basal area of tree species recorded from ground survey. Abbreviations for the beetle species are as in Fig. 2 and those for the tree species are as in Fig. 3.

Figure 5:

the first and the third axis of the ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative basal area of tree species recorded from ground survey. Abbreviations for the beetle species are as in Fig. 2 and those for the tree species are as in Fig. 3.

Figure 5:

the first and the third axis of the ordination diagram resulting from a redundancy analysis between the beetle assemblage and the relative basal area of tree species recorded from ground survey. Abbreviations for the beetle species are as in Fig. 2 and those for the tree species are as in Fig. 3.

Positions of some beetle species also markedly differ relative to the tree vectors between ordinations based on the two tree survey techniques. In the ground survey data set, e.g. S. haematopus seems to have an affinity for both white and black spruce (Figs 4 and 5), while this beetle seems to be mostly restricted to white spruce stands in the ordination based on AVI (Fig. 3). Associations between presence of tamarack and A. gratiosum and P. mannerheimii are more obvious for ordinations based on ground survey data than for those from the photography-based AVI. Similarly, the association between P. brevicornis and spruce-fir forest and between Carabus chamissonis and the mixed forest (Fig. 4) is more obscure in the AVI ordination (Fig. 3).

Accuracy of inventory data derived from aerial photography

Differences in local tree species composition are apparent between the AVI and the ground data sets. Table 2, derived from the confusion matrices of Appendix 1, shows the proportion of sites in which the most abundant tree species were overestimated, underestimated or accurately predicted from air photo-interpretation when compared to ground truthed data based on the ground survey. For the relative amount of both aspen and white spruce, predictions based on interpretation of aerial photography are accurate in 70% of the sites; estimates are more often accurate for balsam poplar and black spruce. Abundance of aspen and white spruce, the most common tree species, are in general more frequently overestimated than underestimated when assessed by air photo, but the reverse is true for balsam poplar and black spruce. Thus, the AVI data tend to underestimate the abundance of these two tree species, which are rarer in the forest.

Table 2:

accuracy estimation of relative canopy cover by AVI for the four most abundant tree species when compared to 198 ground surveyed sites

 Populus tremuloides Populus balsamifera Populus glauca Populus mariana 
 n sites n sites n sites n sites 
Accurate 138 70 170 86 137 69 148 75 
Overestimated 45 23 36 18 
Underestimated 15 26 13 25 13 42 21 
 Populus tremuloides Populus balsamifera Populus glauca Populus mariana 
 n sites n sites n sites n sites 
Accurate 138 70 170 86 137 69 148 75 
Overestimated 45 23 36 18 
Underestimated 15 26 13 25 13 42 21 

In the sites where aspen is overestimated in AVI, balsam poplar and white spruce are both significantly underestimated (Fig. 6a). This same figure shows that in sites where aspen is underestimated, white spruce is significantly overestimated. AVI data also significantly underestimate the abundance of black spruce in sites where white spruce is overestimated, and relative abundance of P. mariana and aspen is overestimated in sites where white spruce is underestimated (Fig. 6b). In cases where black spruce is underestimated, white spruce is significantly overestimated and in the less frequent cases where black spruce is overestimated, white spruce is significantly underestimated (Fig. 6c). The low number of sites with overestimation of balsam poplar in the confusion matrix does not allow the detection of trends in species bias. However, when this species is underestimated, abundance of aspen is significantly overestimated in the provincial data (Fig. 6d). Accuracy assessments for the four other tree species were not calculated because these trees are relatively rare on the landscape and the fact that we considered such a wide interval for accurate classification (±20% on each side of the value) makes any emerging patterns suspicious.

Figure 6:

average percentage (±SE) by tree species in misclassified sites estimated by air photo-interpretation (dark gray) and ground survey (light gray). * indicates a significant (P < 0.05) difference between provincial inventory and ground survey. Refer to Fig. 3 for abbreviations of tree species. Over: overestimated, under: underestimated. Pannels a, b, c and d identify the figure for each tree species.

Figure 6:

average percentage (±SE) by tree species in misclassified sites estimated by air photo-interpretation (dark gray) and ground survey (light gray). * indicates a significant (P < 0.05) difference between provincial inventory and ground survey. Refer to Fig. 3 for abbreviations of tree species. Over: overestimated, under: underestimated. Pannels a, b, c and d identify the figure for each tree species.

DISCUSSION

Source of vegetation data and depiction of biodiversity

Forest vegetation patterns reflected in both the Canadian ecological land classification system (ecosite), as applied in west-central Alberta (Beckingham et al. 1996), and the Alberta provincial forest inventory (AVI) data derived from aerial photography are clearly congruent with spatial patterns of ground beetle biodiversity (Fig. 1). These sources of data about vegetation are broadly available in developed countries for use in landscape scale management of forested land. Analysis presented here suggests that they can provide a quantitative basis for development of coarse filter biodiversity conservation strategies and help guide efficient and effective biodiversity monitoring efforts.

This work reveals a net gain in predictive power (12%) when ecosite classification data are used, instead of maps of relative canopy cover available from AVI. Because the ecosite classification system integrates geospatial information about parent material, slope, aspect, elevation, soils, nutrients, moisture regime and vegetation (Beckingham et al. 1999), beetle habitat is better modeled. This illustrates the advantage of using a composite geo-referenced database for modeling biodiversity patterns as suggested by Pressey (2004). It is imperative to include many relevant habitat variables in order to develop a surrogate system that adequately represents landscape biodiversity. Quality of input data about the taxa being modeled also contributes to the overall biodiversity predictive power of any surrogate system. Thus, it is of much interest to explore these relationships for other groups in which distribution and abundance varies at the stand level.

We found that accuracy of relative canopy cover estimated through interpretation of aerial photography varies between 70 and 86% (Table 2) and that we gained an additional 10% accuracy using tree species mix obtained from direct field survey in predicting ground beetle assemblages over data interpreted from aerial photographs (Fig. 1). This suggests that improvement in detection and estimation of tree species abundance on a landbase is crucial to the development of adequate biodiversity surrogates based on vegetation as small inaccuracies in tree species mixes may significantly impact the final classification (Nielsen et al. 1999).

Given the linear relation between tree basal area and canopy cover (Spurr 1960), a similar amount of variance in beetle assemblages should be explained by both data sets. Errors in the identification and estimation of tree species canopy cover inherent in air photo-interpretation likely drives the discrepancy observed in these results. The AVI did nonetheless perform reasonably well at explaining the multivariate structure of ground beetle assemblages; 78.3% of the variance explained was congruent with what was explained by the data from ground surveys (18 of 23%).

Although similar patterns of association between vegetation and beetle diversity emerged using either the ground survey of trees or the AVI (Figs 3 and 4), some differences were apparent. These differences may affect applicability of these data for conservation and biodiversity monitoring efforts. For example, in our study landscape, strong association between P. mannerheimii and tamarack is obvious using field survey data (Figs 4 and 5) but more obscure when the association is modeled from the AVI data (Fig. 3). It is well understood that P. mannerheimii is characteristic of productive wet sites dominated by Picea and Larix (Larochelle and Larivière 2003). Furthermore, this carabid is recognized in both North America and Fennoscandia as being an uncommon faunal element, being locally restricted by a narrow microhabitat requirement to old wet forests and fire skips (Gandhi et al. 2001; Haila et al. 1994; Niemelä 1997; Niemelä et al. 1992; Paquin 2008). The ecosite classification map captures the distinct habitat occupied by P. mannerheimii, even though the tree canopy cover estimated in AVI is included in developing the ecosite classification. Thus, the ecosite classification system increases the relevance of AVI data for prediction of biodiversity, reinforcing the use of composite geo-referenced databases for biodiversity mapping. Inclusion of many types of variables in the ecosite description compensates for other variables assessed with lower confidence.

Another interesting case is the contrast in associations of S. haematopus suggested by alternative descriptions of forest vegetation. The AVI data (Fig. 3) suggest a strong association with white spruce, although this species seems to occupy the zone of habitat overlap between white spruce and black spruce in the ordination based on the ground survey data (Figs 4 and 5). Comparison of the data sets for the two spruce species (Appendix 1) reveals that white spruce is more often overestimated in the AVI while black spruce was more often underestimated when compared to the ground survey data (Table 2). This suggests that confusion between the two spruce species in the AVI data leads to an overestimation of the strength for the association between S. haematopus and white spruce and an underestimation of the association between this beetle species and black spruce. Although seemingly consistent with published accounts of S. haematopus being associated with drier conifer sites (Larochelle and Larivière 2003; Lindroth 1966), results presented here show that many upland stands of black spruce are mistakenly identified as white spruce (see also Spurr 1960). Thus, the association between S. haematopus and black spruce does not emerge in analyses based on the provincial inventory data.

There is a similar problem with estimation for two Populus species, with balsam poplar often misinterpreted as aspen in data from aerial photography (Table 2 and Fig. 6a and d; Spurr 1960). Such misclassification also affects the depiction of associations between the forest vegetation and the ground beetle fauna. In the ordination based on the ground survey (Fig. 4), the vector for balsam poplar is associated with A. retractum and P. decentis and the vector for aspen is closer to the vectors for C. ingratus and P. adstrictus. This is in accordance with the published information about habitat use in these species; i.e. aspen, C. ingratus and P. adstrictus occur generally on dryer grounds than balsam poplar, A. retractum, and P. decentis (Burns and Honkala 1990; Larochelle and Larivière 2003). However, in the ordination based on the AVI (Fig. 3), both Populus species are associated with A. retractum and P. decentis. Clearly, better differentiation of the fauna associated with each of these two deciduous tree species is achieved in the model based on ground surveys of stand composition (Figs 3–5).

Relative position of vectors for less abundant tree species, such as lodgepole pine and balsam fir, and the beetles P. brevicornis and C. chamissonis, differ greatly between the ground and the AVI-based model (Figs 3 and 4). The methodology employed here does not allow us to effectively detect bias in air photo-interpretation for these less abundant tree species. However, the discrepancy between the ecology modeled from ground survey and the air photo-data for the less common tree species implies that detection and estimation of less abundant tree species from air photo-interpretation does not concur with what was found on the ground. We suggest that accurate detection of surrogate classes that are rare (in this case, less than 2% of the total basal area of all tree sampled by ground survey) on forested landscapes is an important part of an effective approach to employing forest vegetation as a surrogate system for biodiversity. This implies that accurate detection and location of these less abundant surrogate classes provide opportunity to manage landscapes in a manner that should conserve the unique biodiversity gradients associated with these rare trees and, as a consequence, the associated biota.

The use of ground data better represents the finer scale mixture of both deciduous and coniferous components than AVI data. This is based on, e.g. by the narrower angle between the white spruce vector and the balsam poplar and aspen vectors in the field survey model (Fig. 4) compared to that seen in the model based on the AVI data (Fig. 3). Direct comparisons of relative tree species composition confirm that air photo-interpreters tend to rarely assign intermediate relative canopy cover values (Appendix 1), perhaps reflecting a former provincial policy to manage forest landscapes in terms of either deciduous or coniferous stands. Figure 6 also shows that deciduous species are underestimated in sites where coniferous species are overestimated and vice versa. Landscape management based on this sort of air photo-interpretation will indeed tend to ‘unmix’ the mixedwood forest (Magnussen 1997) and thereby affect the distribution and relative abundances of species that depend upon the natural scale of the mix.

Accuracy of inventory data interpreted from aerial photography

For the four tree species common enough for investigation, overall agreement between AVI and ground survey varied between 70 and 86% (Table 2). These species composition accuracy values are comparable to those reported by Fent et al. (1995) for 1:500 photographic images of the Alberta mixedwood forest and correspond to the accuracy range of the best Canadian forest inventory maps (Leckie and Gillis 1995). However, only balsam poplar achieved the benchmark of 85% mapping accuracy generally accepted in the remote sensing literature for validating maps (Anderson 1971; Thompson et al. (2007); Wulder et al. 2006) demonstrated that such errors in estimation of relative tree canopy cover in Ontario forest inventory maps did affect temporal projection of wood supply and introduce serious inaccuracies in models and maps of wildlife habitat. Our results also suggest that inaccuracies in forest inventory obtained from air photo-interpretation lead to erroneous modeling of ground beetle habitat in the mixedwood boreal forest of northwestern Alberta and may impede the use of remotely acquired data for biodiversity conservation.

The inaccuracies in ground beetle habitat models originating from errors in canopy cover estimation may partly be mitigated and bolstered by using a combination of multiple geo-referenced layers representing many habitat variables. From the net gain of 12% in overall predictive power using ecosite classification data, 5% were gained by describing patterns detected by ground survey of trees but undetected by air photo-interpretation (Fig. 1). Complementing the relative canopy cover derived from air photo with other land variables to produce an ecosystem classification map (ecosite; Nielsen et al. 1999) allowed us to capture some of the biodiversity patterns related to canopy cover that were undetected with the sole use of canopy cover from air photo-interpretation. Despite the fact that ecosite classification may enhance our ability to model biodiversity patterns, the use of accurate data is essential to gain as much predictive power as possible and ultimately being able to include the most complete information on biodiversity in conservation on managed forest lands. Technological developments in computer assisted analysis of remotely acquired images promise improvement in tree species differentiation and relative amount estimation, as well as spatial–temporal consistency of the generated data set, and this will improve management for both biodiversity value and fiber resources.

Provincial forest inventory maps currently used in forest management weakly correspond to landscape patterns observed in ground beetle assemblages. Use of such map data as a management surrogate for strategic implementation of biodiversity conservation on the landscape allocated for extensive harvest provides only minimal knowledge to represent regional biodiversity. Combining these maps with other biophysical geo-referenced databases to produce ecologically based land classification improved our ability to detect biodiversity patterns and should improve our strategic framework for managing biodiversity patterns.

Inaccuracies in the databases derived solely from remotely sensed images lead to biases in biodiversity habitat models, therefore limiting accurate spatial and temporal assessment of regional biodiversity patterns. Despite the general concordance of the ecological associations modeled from field or remotely sensed habitat characteristics, we found that divergence in estimates of relative canopy cover confused detection of ecological associations between distributions of beetle and tree species.

CONCLUSION

Implementation of effective conservation strategies for the extensively harvest portion of the boreal forest must start with an accurate, high resolution, wide ranging and spatially explicit knowledge of relevant landscape habitat parameters. Fieldwork to correlate biodiversity patterns with remotely sensed environmental parameters and assess the accuracy of the habitat maps will support development of biodiversity surrogates to meet conservation goals, as we show in this study. In the absence of such surrogates, resource constraints for conservation efforts presently mean that there is little effective protection of ‘biodiversity’ (as opposed, perhaps, to protection of a handful of charismatic species) for large areas of extensively managed forest. In fact, the Canadian ecosystem classification system has the potential to serve as framework for reporting biodiversity conservation requirements to public and governmental agencies. The possibility of including biodiversity along with other land values, such as ecosystem services and local community values (Naidoo et al. 2008), suggests that spatial databases can be developed as powerful decision tools for regional land management.

FUNDING

The work was supported financially by our industrial forestry partners, Canadian Forest Products, Ltd., Daishowa-Marubeni International, Ltd. and Manning Diversified Forest Products, Ltd. as well as Alberta Sustainable Resource Development, the Sustainable Forest Management Network, the Canadian Forest Service and the Natural Sciences and Engineering Research Council of Canada (NSERC).

We thank D. Lysick, D. Jensen, B. Shaughnessy, K. Cryer, I. Phillips, D. Hartley, Z. Bergeron as well as the EMEND field crew for their help in the field and in the laboratory. Discussions with D. W. Langor, M. Koivula, T. T. Work and J. Jacobs greatly contributed to the development of ideas presented in this paper.

Conflict of interest statement. None declared.

APPENDIX 1

graphic

References

Alberta Environmental Protection
Alberta Timber Harvest Planning and Operating Ground Rules
 , 
1994
Edmonton, Canada
Alberta Government
 
Pub. No.: Ref. 71
Anderson
JR
Land use classification schemes used in selected recent geographic applications of remote sensing
Photogramm Eng Remote Sens
 , 
1971
, vol. 
37
 (pg. 
379
-
87
)
Avery
TA
Berlin
GL
Fundamentals of Remote Sensing and Air Photo Interpretation
 , 
2003
6th edn
New York
Prentice Hall
 
540 pp
Bailey
RG
Hogg
HC
A world ecoregion map for resources reporting
Environ Conerv
 , 
1986
, vol. 
13
 (pg. 
195
-
202
)
Barbaro
L
Rossi
J-P
Vetillard
F
, et al.  . 
The spatial distribution of birds and carabid beetles in pine plantation forests: the role of landscape composition and structure
J Biogeogr
 , 
2007
, vol. 
34
 (pg. 
652
-
64
)
Beckingham
JD
Corns
IGW
Archibald
JH
Field guide to ecosites of west-central Alberta
Special report 9
 , 
1996
Edmonton, Canada
Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre
 
540 pp
Beckingham
JD
Desilets
M
Nielsen
D
, et al.  . 
Multi-layered and statistically based ecosystem mapping: the de facto standard for land resource planning in the 21st century
Earth Obs Mag
 , 
1999
, vol. 
8
 (pg. 
10
-
3
)
Blanchet
FG
Legendre
P
Borcard
D
Forward selection of explanatory variables
Ecology
 , 
2008
, vol. 
89
 (pg. 
2623
-
32
)
Borcard
D
Legendre
P
Drapeau
P
Partialling out the spatial component of ecological variation
Ecology
 , 
1992
, vol. 
73
 (pg. 
1045
-
55
)
Bousquet
Y
Checklist of Beetles of Canada and Alaska. Publication 1861/E
 , 
1991
 
. Ottawa, Canada: Research Branch, Agriculture Canada. 430 pp
Burns
RM
Honkala
BH
Silvics of North America: 1. Conifers; 2. Hardwoods. Agriculture Handbook 654
 , 
1990
Washington, DC
U.S. Department of Agriculture, Forest Service
Congalton
RG
Accuracy assessment and validation of remotely sensed and other spatial information
Int J Wildland Fire
 , 
2001
, vol. 
10
 (pg. 
321
-
8
)
Czaplewski
RL
Wulder
MA
Franklin
SE
Accuracy assessment of maps of forest condition
Remote Sensing of Forest Environment
 , 
2003
Norwell, MA
Kluwer Academic Publishers
(pg. 
115
-
40
)
Dray
S
Legendre
P
Blanchet
FG
packfor: Forward Selection with Permutation (Canoco p.46)
 , 
2009
 
R package version 0.0-7/r58. http://R-Forge.R-project.org/projects/sedar/ (November 2011, date last accessed)
Duro
DC
Coops
NC
Wulder
MA
, et al.  . 
Development of a large area biodiversity monitoring system driven by remote sensing
Prog Phys Geogr
 , 
2007
, vol. 
31
 (pg. 
235
-
60
)
Eyre
MD
Luff
ML
Ground beetle species (Coleoptera, Carabidae) associations with land cover variables in northern England and southern Scotland
Ecography
 , 
2004
, vol. 
27
 (pg. 
417
-
26
)
Fent
L
Hall
RJ
Nesby
RK
Aerial films for forest inventory: optimizing film parameters
Photogramm Eng Remote Sens
 , 
1995
, vol. 
61
 (pg. 
281
-
9
)
Ferrier
S
Mapping spatial pattern in biodiversity for regional conservation planning: where to go from Here?
Syst Biol
 , 
2002
, vol. 
51
 (pg. 
331
-
63
)
Foody
GM
Status of land cover accuracy assessment
Remote Sens Environ
 , 
2002
, vol. 
80
 (pg. 
185
-
201
)
Franklin
JF
Preserving biodiversity: species, ecosystems, or landscapes?
Ecol Appl
 , 
1993
, vol. 
3
 (pg. 
202
-
5
)
Gandhi
KJK
Spence
JR
Langor
DW
, et al.  . 
Fire residuals as habitat reserve for epigaeic beetles (Coleoptera: Carabidae and Staphylinidae)
Biol Conserv
 , 
2001
, vol. 
102
 (pg. 
131
-
41
)
Haila
Y
Hanski
IK
Niemelä
J
, et al.  . 
Forestry and boreal fauna: matching management with natural forest dynamics
Ann Zool Fenn
 , 
1994
, vol. 
31
 (pg. 
187
-
202
)
Hall
RJ
Wulder
MA
Franklin
SE
The role of aerial photographs in forestry remote sensing image analysis
Remote Sensing of Forest Environment
 , 
2003
Norwell, MA
Kluwer Academic Publishers
(pg. 
45
-
75
)
Kerr
JT
Southwood
TRE
Cihlar
J
Remotely sensed habitat diversity predicts butterfly species richness and community similarity in Canada
Proc Natl Acad Sci U S A
 , 
2001
, vol. 
98
 (pg. 
11365
-
70
)
Larochelle
A
Larivière
M-C
A Natural History of the Ground-Beetles (Coleoptera: Carabidae) of America North of Mexico
 , 
2003
Sofia, Bulgaria
Pensoft
Leckie
DG
Gillis
MD
Forest inventory in Canada with emphasis on map production
For Chron
 , 
1995
, vol. 
71
 (pg. 
74
-
88
)
Legendre
P
Durand
S
rdaTest: Canonical Redundancy Analysis
 , 
2010
 
R package version 1.7-1. http://www.bio.umontreal.ca/legendre/indexEn.html (November 2011, date last accessed)
Legendre
P
Gallagher
E
Ecologically meaningful transformations for ordination of species data
Oecologia
 , 
2001
, vol. 
129
 (pg. 
271
-
80
)
Legendre
P
Legendre
L
Numerical Ecology
 , 
1998
2nd English edn
Amsterdam, The Netherlands
Elsevier Science B.V
Leyequien
E
Verrelst
J
Slot
M
, et al.  . 
Capturing the fugitive: applying remote sensing to terrestrial animal distribution and diversity
Int J Appl Earth Obs Geoinf
 , 
2007
, vol. 
9
 (pg. 
1
-
20
)
Lindroth
CH
The ground-beetles of Canada and Alaska
Opusc Entomol
 , 
1961
, vol. 
Supplementum XX
 (pg. 
1
-
200
)
Lindroth
CH
The ground-beetles of Canada and Alaska
Opusc Entomol
 , 
1963
, vol. 
Supplementum XXIV
 (pg. 
201
-
408
)
Lindroth
CH
The ground-beetles of Canada and Alaska
Opusc Entomol
 , 
1966
, vol. 
Supplementum XXIX
 (pg. 
409
-
648
)
Lindroth
CH
The ground-beetles of Canada and Alaska
Opusc Entomol
 , 
1968
, vol. 
Supplementum XXXIII
 (pg. 
649
-
944
)
Lindroth
CH
The ground-beetles of Canada and Alaska
Opusc Entomol
 , 
1969
, vol. 
Supplementum XXXIV
 (pg. 
945
-
1192
)
Magnussen
S
A method for enhancing tree species proportions from aerial photos
For Chron
 , 
1997
, vol. 
73
 (pg. 
479
-
87
)
Margules
CR
Pressey
RL
Systematic conservation planning
Nature
 , 
2000
, vol. 
405
 (pg. 
243
-
53
)
Marshall
IB
Scott Smith
CA
Shelby
CJ
A national framework for monitoring and reporting on environmental sustainability in Canada
Environ Monit Assess
 , 
1996
, vol. 
39
 (pg. 
25
-
38
)
McMullan-Fisher
SJM
Kirpatrick
JB
May
TW
, et al.  . 
Surrogates for macrofungi and mosses in reservation planning
Conserv Biol
 , 
2009
, vol. 
24
 (pg. 
730
-
6
)
Morgan
JL
Gergel
SE
Coops
NC
AerialpPhotography: a rapidly evolving tool for ecological management
Bioscience
 , 
2010
, vol. 
60
 (pg. 
47
-
59
)
Moss
EH
Flora of Alberta
 , 
1983
2nd edn
, revised by John G. Packer. Toronto, Canada
University of Toronto Press
Müller
J
Brandl
R
Assessing biodiversity by remote sensing in mountainous terrain: the potential of LiDAR to predict forest beetle assemblages
J Appl Ecol
 , 
2009
, vol. 
46
 (pg. 
897
-
905
)
Nadeau
LB
Li
C
Hans
H
Ecosystem mapping in the Lower Foothills Subregion of Alberta: application of fuzzy logic
For Chron
 , 
2004
, vol. 
80
 (pg. 
359
-
65
)
Nagendra
H
Using remote sensing to assess biodiversity
Int J Remote Sens
 , 
2001
, vol. 
22
 (pg. 
2377
-
400
)
Naidoo
R
Balmford
A
Costanza
R
, et al.  . 
Global mapping of ecosystem services and conservation priorities
Proc Natl Acad Sci U S A
 , 
2008
, vol. 
105
 (pg. 
9495
-
500
)
Nielsen
D
Coenen
V
Beckingham
JD
Ecosite and Ecosection Mapping of the Canfor, Hines Creek Operating Area (P1 South, P2, and P11 Forest Management Units)
 , 
1999
Edmonton, Canada
Geographic Dynamics Corp.
 
, Ref. # 1999033
Niemelä
J
Invertebrates and boreal forest management
Conserv Biol
 , 
1997
, vol. 
11
 (pg. 
601
-
10
)
Niemelä
J
Spence
JR
Spence
DH
Habitat association and seasonal activity of ground-beetles (Coleoptera: Carabidae) in central Alberta
Can Entomol
 , 
1992
, vol. 
124
 (pg. 
521
-
40
)
Noss
RF
Indicators for monitoring biodiversity: a hierarchical approach
Conserv Biol
 , 
1990
, vol. 
4
 (pg. 
355
-
64
)
Oksanen
J
Blanchet
FG
Kindt
R
, et al.  . 
vegan: Community Ecology Package
 , 
2011
 
R package version 1.17-10. http://vegan.r-forge.r-project.org/ (November 2011, date last accessed)
Olson
DM
Dinerstein
E
Wikramanayake
ED
, et al.  . 
Terrestrial ecoregions of the world: a new map of life on earth
Bioscience
 , 
2001
, vol. 
51
 (pg. 
933
-
8
)
Paine
DP
Kiser
JD
Aerial photography and image interpretation
 , 
2003
2nd edn
Hoboken, NJ
John Wiley & Sons Inc
Paquin
P
Carabid beetle (Coleoptera: Carabidae) diversity in the black spruce succession of Eastern Canada
Biol Conserv
 , 
2008
, vol. 
141
 (pg. 
261
-
75
)
Pressey
RL
Conservation planning and biodiversity: assembling the best data for the job
Conserv Biol
 , 
2004
, vol. 
18
 (pg. 
1677
-
81
)
R Development Core Team, (2011)
R: A Language and Environment for Statistical Computing
 
Vienna, Austria
R Foundation for Statistical Computing
 
ISBN 3-900051-07-0. http://www.R-project.org/ (November 2011, date last accessed)
Richards
JA
Jia
X
Remote Sensing Digital Image Analysis
 , 
2006
4th edn
Berlin, Germany
Springer
Rodrigez
JP
Brotons
L
Bustamante
J
, et al.  . 
The application of predictive modeling of species distribution to biodiversity conservation
Divers Distrib
 , 
2007
, vol. 
13
 (pg. 
243
-
51
)
Spence
JR
Langor
DW
Arthropods as ecological indicators of sustainability in Canadian forests
For Chron
 , 
2006
, vol. 
82
 (pg. 
344
-
50
)
Spence
JR
Langor
DW
Jacobs
JM
, et al.  . 
Conservation of forest-dwelling arthropod species: simultaneous management of many heterogeneous risks
Can Entomol
 , 
2008
, vol. 
140
 (pg. 
510
-
25
)
Spence
JR
Niemelä
JK
Sampling carabid assemblages with pitfall traps: the madness and the method
Can Entomol
 , 
1994
, vol. 
126
 (pg. 
881
-
94
)
Spence
JR
Volney
WJA
Lieffers
VJ
, et al.  . 
Veeman
TS
Smith
DW
Purdy
BG
Salkie
FJ
Larkin
GA
The Alberta EMEND Project: recipe and cooks' argument
Science and Practice: Sustaining the Boreal Forest. Edmonton, Canada: SFM Network, 14–17 February 1999
 , 
1999
(pg. 
583
-
90
)
Spurr
SH
Photogrammetry and Photo-interpretation
 , 
1960
2nd edn
New York, NY
The Ronald Press Company
Stehman
SV
Czaplewski
RL
Design and analysis for thematic map accuracy assessment: fundamental principles
Remote Sens Environ
 , 
1998
, vol. 
64
 (pg. 
331
-
44
)
Thompson
ID
Maher
SC
Rouillard
DP
, et al.  . 
Accuracy of forest inventory mapping: some implications for boreal forest management
For Ecol Manag
 , 
2007
, vol. 
252
 (pg. 
208
-
21
)
Turner
W
Spector
S
Gardiner
N
, et al.  . 
Remote sensing for biodiversity science and conservation
Trends Ecol Evol
 , 
2003
, vol. 
18
 (pg. 
306
-
14
)
Wilson
EO
Biodiversity
 , 
1988
Washington, DC
National Academy Press
Wulder
MA
Franklin
SE
White
JC
, et al.  . 
An accuracy assessment framework for large-area land cover classification products derived from medium-resolution satellite data
Int J Remote Sens
 , 
2006
, vol. 
27
 (pg. 
663
-
83
)