The figure below outlines the general procedure the ABMI uses to produce species models and, later, use them to calculate intactness for species, groups of species, and overall biodiversity. Field surveys provide data on the relative abundances of species and habitat elements at ABMI sites. We use province-wide maps of vegetation and human footprint created by the ABMI to describe the landbase at each site.

Statistical analyses generate models of how each species’ abundance relates to the vegetation and human footprint at the sites. These models are summarized and applied in different ways to generate several products, including estimates of relative abundance of each species in each vegetation or human footprint type, and maps of the predicted current distribution of the species across the province. The models are also applied to “reference” vegetation maps, in which human footprint has been replaced, or “back-filled”, with an estimate of pre-disturbance vegetation. Intactness of a species compares the predicted current abundance of the species to the abundance predicted under the reference condition with no human footprint. Intactness of a group of species, such as old-forest plants or human-associated birds, or of a larger taxonomic group, such as all mites, is calculated by averaging the intactness across all species in the group. Overall biodiversity intactness is an average of the intactness values for the larger taxonomic groups.

Data partners

The ABMI's program and resulting data and information products are made stronger through partnership and collaboration with various organizations and individuals, many of whom contribute important data to the ABMI. For example:

• For the ABMI’s bird analyses, ABMI data are combined with data from the Boreal Avian Modelling Project (BAM). Within Alberta, the BAM database is a compilation of data from the Breeding Bird Survey (BBS), Breeding Bird Atlases, Environment and Climate Change Canada (ECCC), the Bioacoustic Unit at the University of Alberta, other monitoring projects, and short-term research projects.
• ABMI human footprint data are enhanced by a longstanding partnership with the Government of Alberta / Alberta Environment and Parks (AEP) through the Alberta Human Footprint Monitoring Program (AHFMP). This partnership allows, for example, additional geospatial layers produced or maintained by the Government of Alberta to be incorporated into the ABMI's Human Footprint Inventory.
• Environment and Climate Change Canada supports the ABMI's work in a variety of ways. In 2019, acoustic experts from ECCC, the ABMI, the Bioacoustic Unit, and BAM came together to develop the next phase of acoustic data management for Canada, which has come to be called the Open Data Initiative (ODI). This initiative, funded by ECCC, involves further functional development of the ABMI's WildTrax platform, and increased standardization and data sharing across data management platforms like WildTrax and NatureCounts.
• The Caribou Monitoring Unit (CMU), itself the product of a large inter-organizational collaboration, provides data related to its leading role supporting woodland caribou recovery in Canada.

These examples are representative but not exhaustive. We are grateful to all of our partners and collaborators for their ongoing operational, financial, and scientific support.

Summary: Species–habitat models relate the abundance of target species to habitat attributes. They can be used to predict species abundance for any point in time or space for which habitat information is available.

Modelling the relationship of species to habitat is at the heart of many ABMI analyses, including direct descriptions of those habitat relationships, mapping, and species intactness. “Habitat” is used in a broad sense, including:

• Human footprint
• Descriptions of the natural ecosystems—vegetation (stand types by age classes) in forested areas, or soils in prairie areas
• Climate and spatial variables (latitude, longitude), and
• Footprint surrounding a site. (The effects of footprint surrounding a site are described below, but these are not currently used in our products, for reasons we discuss.)

We use a flexible analysis for the habitat modelling, which can be adapted to address specific questions, such as the effects of particular industrial sectors (more on that in its own section below).

Relative abundance by human footprint and ecosystem type

Modelling Species: The first, most basic step of the analysis is to estimate the relative abundance of a species in different human footprint types and ecosystem types. Because most ABMI sites cover several habitat types, we use a multiple regression approach to separate the effect of each type. The main footprint and ecosystem variables used in the models and modelling procedure are the same among taxa, but details of the modelling differ slightly due to different sampling methods for the taxa and habitat elements.

Note: The following descriptions reference data collection procedures outlined in the ABMI's Terrestrial Field Data Collection Protocols.

Vascular plants and mites: The data for both taxa are treated as occurrences in 4 quadrats (plants) or soil samples (mites) in the 1-ha central area of each site, using a logit-link binomial distribution with 4 trials. Vegetation and human footprint variables are summarized for the 1 ha central area.

Bryophytes and lichens: The data since 2009 are similar to the vascular plant data, and are also analyzed as 0-4 quadrat occurrences per site. Before 2009, these taxa were surveyed with a more complex design that stratified by microhabitat across the central 1 ha plot. Those data are reduced to simple presence/absence at the site. The pre- and post-2009 data are analysed together, using a logit-link binomial distribution with 1 (pre-2009) or 4 (post-2009) trials, and an additional factor for old versus new protocol. The protocol factor adjusts for the different area of each occurrence (1 ha versus 50 m x 50 m) and different search protocol.

Mammals: Mammals were originally surveyed using snow track transects with a triangle of 3 km per side. The protocol was changed to a 10 km linear transect in 2005. In both methods, occurrences of mammal species were recorded separately for 1 km segments. Mammals are analysed as presence/absence on each 1 km segment, with vegetation and human footprint summarized in a 250 m buffer around each segment. Adjustments are made to estimated standard errors to compensate for the dependence of segments along the same transect, and transects are resampled together in bootstrapping. A factor is included in the analysis to account for the change in protocol from triangles to transects. Additionally, tracking data are directly affected by days-since-snow when the survey was conducted. A quadratic relationship with days-since-snow is included in the models to account for this.

In 2015, the ABMI switched from snow track transects to cameras for monitoring mammals.

Birds: The ABMI bird data are modelled as the number of occurrences in the 9 point counts at a site, using a logit-link binomial distribution with 9 trials. The raw data for the bird surveys records the number of individuals of a species at a point count station, but we reduce this to presence/absence at each station, because most values are 0 or 1, and repeatability of the data for multiple individuals at a station is low. Vegetation and human footprint variables are summarized for the total area occupied in 150 m-radius circles around each of the 9 count stations at a site.

The above description applies to analyses of only the bird data collected by the ABMI. However, most bird results presented by the ABMI, including on the website and in intactness analyses, use a larger dataset provided by the Boreal Avian Modelling (BAM) project*. The BAM bird data includes ABMI data, Breeding Bird Survey data, and datasets collected by Environment Canada and researchers throughout the boreal forest. These analyses are more detailed because the combined dataset for birds is richer than that for other taxa. The models produced, however, are the same as those produced for other taxa, and they are used in the same way to produce our various products.

The various data sources were standardized using statistical offsets (see Sólymos et al. (2014) for additional references and more detailed description of the bird modelling).

* Note: Within Alberta, the BAM database is a compilation of data from the Breeding Bird Survey (BBS), Breeding Bird Atlases, Environment and Climate Change Canada (ECCC), the Bioacoustic Unit at the University of Alberta, other monitoring projects, and short-term research projects. In particular, the ABMI, BAM, and ECCC have enjoyed a long and mutually beneficial relationship during which both data and ideas have been shared. The bird data and information reported by the ABMI substantially benefits from the aggregate dataset of these organizations, as well as the efforts and expertise of the collaborating scientists.

Habitat elements – trees, snags and logs: Counts of these structures, divided into classes by diameter and species groups, are analysed with log-link negative binomial count models, which can capture the highly aggregated nature of some structures (a few high counts, many zeroes).
 
Predicted counts from these models can be converted into densities (/ha) and then into basal areas of trees or snags (m2/ha) or volumes of downed wood (m3/ha), using quadratic mean diameters of the size classes and the known sampling areas or transect lengths.

Habitat elements – cover layers: Canopy, shrub, etc. cover layers are analysed as a logit-linked binomial variable, as this produced more stable models than alternatives such as beta distributions or other transformations.

Habitat elements – miscellaneous: pH is modelled as a log-normal distribution. Organic depth and soil carbon have a normal error distribution.

Additional variation due to climate, geographic location and surrounding human footprint

The analyses in the previous section provide estimates of species’ relative abundances in human footprint and ecosystem types that are the same throughout the analysis zone. In addition to varying due to these factors, we know that many species also vary geographically and/or along climate gradients. Species may also be affected by the amount of footprint in an area around a site, not just the footprint at the site. For example, a species that uses native vegetation may be less common at a site with native vegetation if it is surrounded by human footprint than if it is part of a larger area of native vegetation.

The estimates of abundance by human footprint and ecosystem type are first used to predict a species’ abundance at each site. These are then used as offsets in models to estimate the residual effects of geographic location, climate and surrounding human footprint. We use 14 models with different individual or combined climate variables, such as mean annual temperature, potential evapotranspiration, etc. An additional 14 models add latitude and longitude, and a further 14 models add latitude, longitude and latitude*longitude. The best of these 42 models of residual effects of climate + geographic location is selected using BIC.

The climate and/or spatial variables from this best model are then used as covariates in a set of 7 models estimating the residual effects of human footprint in a 1 km2 unit surrounding each site.
 
The 7 models use different ways of grouping the human footprint types. These 7 models of surrounding-footprint effects, including the best climate and spatial covariates, are combined using BIC-weighted model-averaging. Currently, however, we are not using the surrounding- footprint effects in the predictions used to generate intactness estimates, maps and industrial sector effects. Confounding of surrounding footprint with footprint at the site, climate variables and geographic location make it difficult to reliably estimate the incremental effects of surrounding footprint for all species.

Using sites with more than one visit

Some sites have been surveyed more than once. These repeat samples are included in intactness models but weighted by the number of repeats (1/n; in other words, each site has the same overall weighting even if some sites include data from more than one survey). During bootstrap analysis (see Intactness section below), the site is treated as the unit of resampling, such that data from all surveys at a site are resampled together.

As more data from planned re-surveys of sites become available in the future, additional terms will be added to the models to allow the species abundance to differ between initial and subsequent sampling periods. The predicted abundance for the second (or later) measurement cycle could therefore be greater or less than the abundance predicted from habitat and footprint alone. That difference would reflect trends in the species’ abundance beyond those due to footprint changes. Intactness calculations after a second set of samples is begun will include both the footprint effect and any additional changes in the species. This will allow us to attribute changes in species abundances to both the effects of changes in human footprint and other changes unrelated to footprint.

“Targeted” sites

In addition to the 1656 sites on the systematic grid across the province, the ABMI samples many “targeted” sites. These sites are chosen to complement the systematic sites for the purpose of addressing specific short-term questions, and are sampled with the same protocols as the systematic sites. A main objective for targeted sites is to improve sampling coverage along the gradient of human footprint levels, and thus to improve estimation of species-footprint relationships. Targeted sampling is therefore focused on the end of the footprint gradient that is less common in a region: high footprint sites in the boreal and foothills region, and low footprint sites in the grasslands and parklands. Effort is made to find sites satisfying these conditions while also being widely distributed and representative of the ecosystem types in the particular region (although this is not always possible). Some targeted sites are chosen in underrepresented footprint types, such as large industrial sites, or address specific management questions, such as the rate of species recovery in older cutblocks. Because we calculate reference and current abundances of species using complete maps of the region of interest rather than just our sampled sites (see Intactness section below), our intactness results are not affected by the fact that the targeted sites are not a representative sample of the footprint in the region.

Wetland habitat models and intactness

Note: The following descriptions reference data collection procedures outlined in the ABMI's Wetland Field Data Collection Protocols.

In addition to the systematic sites, the ABMI conducts additional surveys targeted at small wetlands with open water. The systematic sites can fall in upland or lowland habitat, but the field surveys cannot be conducted in open water. Open-water wetlands and their adjacent emergent, margin meadow and margin wooded habitats are important distinct ecosystem types that are most efficiently monitored with targeted sites and specific field methods. The ABMI surveys the nearest open-water wetland to each of the systematic sites, with transect-based surveys of plants and aquatic invertebrates.

Wetland habitat models

Analysis of footprint effects on wetland species takes a different approach than analysis with the systematic sites, because the relevant footprint is not just in the wetland itself, but also in the surrounding habitats. Additionally, habitat classification and mapping of wetland types is less developed than for terrestrial habitats, but there are other important ecosystem variables, such as wetland depth or chemistry.

Habitat modeling for wetland species therefore first analyzes three sets of ecosystem covariates: climate and spatial variables, wetland physical and chemical variables (e.g., wetland depth, pH, total nitrogen, etc), and broad surrounding vegetation (North) or soil (South) types. The best sets of these covariates are chosen with a model selection procedure. Using those best covariates, the analysis then examines what effect footprint in the surrounding area has on the species’ abundance. We are currently using a 250 m-wide buffer around the wetland’s water edge to define the surrounding area. Eventually we will change that to a topographically defined catchment for the wetland, which is more relevant biologically. The analysis results in relationships of each species to surrounding footprint types, having factored out the effects of relevant covariates.

We summarize the wetland models with figures showing how the predicted relative abundance of the species changes with increasing levels of each footprint type in the surrounding area, at average values of the relevant covariates, and we tabulate the coefficients of the model for use by others. We cannot currently map wetland species or wetland abundance across the province, because we do not have accurate maps of where the types of wetlands we sample occur, and also because the models include wetland physical and chemical variables that have to be measured in the field for each wetland.

Summary: The Intactness Index provides an estimate of how much a species' abundance has diverged from baseline due to human disturbance. Intactness is summarized in a dedicated "cheat sheet" here.

The ABMI’s intactness index compares the predicted current abundance of each species to its predicted reference abundance. Our reference condition is the current abundance with the effects of human footprint statistically removed using our habitat models (Nielsen et al. 2007). Alternative definitions of reference conditions have been used elsewhere, but these are problematic:

• Abundances in protected areas cannot be used, because protected areas are rarely representative of all ecosystems. In Alberta, many protected areas are in remote or high elevation regions with low productivity.
• Time-zero approaches set the reference condition as abundance in a certain year. However, human footprint has affected parts of Alberta for over a century, and many areas had extensive development when systematic monitoring began in 2007.
• Desired or target abundances of species are sometimes used as references, but these are social values that differ for different people, and it is infeasible to set targets for thousands of species. Using a reference condition based on statistically removing the effects of human footprint overcomes these problems. However, this “de-footprinted” reference condition cannot account for any past changes in the species’ abundance that are not due to local or regional footprints, including historical exploitation, effects of diseases or introduced species that are not associated with footprint, climate change or past effects outside the province for migratory species.

The species intactness index compares the predicted relative abundance of each species across the reporting region to the predicted abundance for that species under zero human footprint in the same region. This measure of intactness is scaled between 0 and 100, with 100 representing current abundance equal to that expected under reference conditions, and 0 representing species abundance as far from reference condition as possible. Both over- and under-abundances are viewed as deviations from intact conditions. The index is estimated as:

Current / Reference × 100%, when Current < Reference, or
Reference / Current × 100%, when Reference < Current

A value of 50%, for example, means that the species is either half as abundant as the reference condition, or twice as abundant. Because the intactness index for individual species decreases from 100% with either downwards or upwards differences from reference conditions, an “increaser” species does not cancel out a “decreaser” species. Instead, both contribute to lowering the average intactness.

Intactness in large regions

The reference and current abundances of a species in a reporting region are calculated by applying the habitat models and models of residual climate/location/surrounding footprint to landbase data from every 1 km2 unit in the region. The landbase data comes from province-wide maps of vegetation, soils and human footprint produced by ABMI. Compiling and updating the human footprint information for the whole province is time-consuming, so regional intactness results usually represent human footprint levels 2 or 3 years prior to reporting.

When we calculate intactness for large diverse regions, such as those used by the Land Use Framework in Alberta, we may have different numbers of species modeled for different parts of the region. Specifically, we have more species modeled in the northern regions of the province, where we have more samples. Simply averaging all species found in a region would bias results towards the northern parts of the region. Instead, we first calculate intactness for each natural sub-region within the region of interest, and then calculate an area-weighted average intactness of the sub-regional values. This procedure treats each area of the region equally, regardless of how many species we have modeled in the area.

Within each subregion, we also exclude species that are predicted to have very low abundances in that subregion. The specific criterion is to exclude any species whose predicted current and reference abundances in the region are both < 1% of the species’ predicted maximum abundance in the province. This criterion excludes modeled species from sub-regions where they are unlikely to occur in reality, while not favouring common species over species that are uncommon everywhere in the province.

Predicted current abundances rather than observed

The intactness index uses the predicted current abundance for each 1 km2 unit when the footprint variables are set to their current levels, rather than directly using the counts at surveyed sites, for three reasons:

• This approach allows us to do the calculations using 1 km2 units that cover the entire area of the region, rather than just the small subsample of the area where we have ABMI sites with the direct counts. That is particularly important for reporting on small subregions.
• Measurement error is large for single counts, and substantial even across multiple sites in a subregion. We do not want to compound true changes in species caused by footprint and the effects of measurement error.
• The reference condition is calculated using regression models with a link function (log or logit). The current condition is also calculated with this link function, to avoid differences due simply to different types of scaling (parallel to the difference between arithmetic and geometric means).

Predicted intactness over time

When we report on changes in intactness over time in a region, we use a subset of 3x7 km areas overlaying ABMI sites (i.e., on a 20 km x 20 km grid) in which human footprint has been mapped from satellite images for most years from 1999 to present. These 3x7 areas are a 5.25% sample of the all regions. Because they are a sample of a region, rather than a complete inventory, they add sampling error to the intactness estimates. Our analyses show that this sampling error adds 2-3% additional error to the intactness estimates for species, which is small compared to the uncertainty in the models generating the current and reference predictions.

The importance of “increaser” species

In areas with low footprint, such as much of northern Alberta, species that increase with human footprint tend to have lower intactness than species that decrease. With 5% footprint in a region, for example, it is rare to have a species that decreases more than 5%. To do so, the species would have to have a strong negative response to footprint and footprint would have to be concentrated in preferred habitat for that species. In contrast, it is more common for some species to increase several-fold as soon as any footprint occurs in a region, because several species associated with footprint are expected to be extremely rare under no-footprint reference conditions – the proportional increase when they show up in footprint is therefore large.
Although species that increase with footprint can be an ecological concern because of negative effects on other species, species that decrease with footprint are often of a greater priority to managers. For this reason, we often report intactness of decreaser and increaser species separately, or highlight decreaser species or groups of species.

Intactness for groups, taxa and overall biodiversity

Intactness for groups of species within a taxon, such as old-forest birds or berry-producing shrubs, is calculated as the average intactness of species in the group. Similarly, intactness for a taxon like bryophytes or mites is the average intactness of all analysed species in that taxon.

Overall biodiversity intactness is the average of the intactness for each of the taxa surveyed by ABMI (birds, mammals, vascular plants, bryophytes, mites, lichens). Each taxon is weighted equally, regardless of how many species it contains.

We considered many other ways of combining species when calculating intactness for species groups, but concluded that a simple mean is the most appropriate. Geometric, harmonic or other types of averages have undesirable properties, including giving excessive weight to individual species with extreme values, which were often the most poorly estimated results (for rare or highly aggregated species). Alternatives that weighted species by the precision of their intactness estimate are biased toward common species. More complicated methods that statistically correct for the expected relationship between abundance and precision could be used to remove this bias with precision-weighted averages. However, the loss of transparency and ease of understanding the results outweigh any possible statistical benefits of such a procedure.

Non-native plants: For non-native species, it does not make sense to calculate an average intactness, because the more rare non-native plant species that are included, the higher the intactness. For example a non-native species that is observed once in a region with 100 sites sampled has an intactness of 99%, and another non-native that is observed at 50 sites has an intactness of 50%. If these intactness values were simply averaged, the first record of a newly invading non-native species would raise the mean intactness, which is not a sensible result. For that matter, potential non-native species that have not yet been detected would have an intactness of 100%, and would raise the average intactness for the overall group even higher. Instead, a different approach is used for non-native plants. We assume that the reference condition is 0 for these species, then calculate a species’ intactness as: SIs = 100% - percent occurrence of the species. Percent occurrence is calculated at the quadrat (50m x 50m) level. For example, a non- native species that occurred in 20% of quadrats would have an intactness value of 80%. We calculate the overall intactness of ALL non-native plants as 100% - percent occurrence of any non-native species on a quadrat. This value is simply the percent of quadrats on which no non- native species was detected in a region.

Bootstrapping and confidence intervals

The sampling distribution of species intactness indices are estimated using bootstrapping, in which the original data are resampled with replacement and the entire analysis repeated 100 times. The bootstrap replicates are used to calculate the median reference condition and confidence intervals (based on percentiles of the 100 resampled values). Bootstrapping is required, rather than an analytical formula, because the current abundances and reference conditions are not independent, and the intactness calculation is complicated for the multi-stage, multi-model approach with different weighting of revisited sites. A blocked bootstrap is used, in which the resampling is done within pre-defined spatial blocks to preserve the spatial structure of the sample design. For mammal snow-tracking, the transect or triangle is used as the unit of resampling. When we are using the 3x7km areas to estimate intactness changes over time, the
 
3x7 areas in the reporting region are also resampled during the bootstrapping to include the additional error due to these areas being a sample of the region, rather than a complete inventory.

Wetland intactness

Because we cannot map wetland species’ current and reference abundances across the province, the procedure for calculating regional intactness is simpler for wetland species than for species in the systematic sites. We predict the current and reference abundance of each wetland species at each sampled wetland – the only places we have the necessary field sampling of the physical and chemical covariates. The current abundance is predicted with footprint in the surrounding 250 m buffer set to the levels currently observed; the reference abundance is predicted with those levels set to 0. As with the systematic sites, we use the predicted current abundance rather than the observed values for each species so that we are not including the effect of measurement error as part of the difference between current and reference values, and to allow an “apples-to-apples” comparison of the same logit-link values. For regional intactness, we simply sum the predicted current and reference abundances of a species for all wetlands sampled in the region, and calculate intactness with the same formula used in the systematic plots.

Summary: Sector effects provide an estimate of the effect of specific sources of human disturbance on the abundance of target species.

We use the habitat models to estimate the effects of individual industrial sectors in a region. We start with a summary of all combinations of habitat types X footprint types in each 1 km2 unit, for both the current and reference (no footprint) conditions. We can then “turn on” only the footprint types associated with one industrial sector and compare the predicted abundances of each species with only that footprint type in the current landbase to the reference landbase. This shows how much the footprint from just that one industrial sector is predicted to have changed the abundance of that species. We do this calculation for the major industrial sectors of agriculture, forestry, energy, rural/urban, and transportation. The analysis gives both the predicted total effect of the industrial sector on each species’ abundance, and the effect per unit area of the industry’s footprint (i.e., how “intensive” the effect of the industry’s footprint is).

The example below shows the estimated percentage change in balsam poplar relative abundance inside areas that have been disturbed by each sector in Alberta's forested (northern) region. In this example, balsam poplar relative abundance is reduced by agricultural footprint, but increased to varying degrees by other footprint types.

The ABMI makes a wealth of information on its methods, data collection protocols, and the research underlying these, available through its Publication Archive.

References for this page:

• ABMI Manual for Species Modelling and Intactness
• ABMI Essentials: Intactness
• ABMI Science Letter: Effects of Industrial Sectors on Species Abundance in Alberta
• ABMI Terrestrial Field Data Collection Protocols
• ABMI Wetland Field Data Collection Protocols
• Nielsen et al. 2007: A New Method to Estimate Species and Biodiversity Intactness Using Empirically Derived Reference Conditions
• Solymos et al. 2014: Development of Predictive Models for Migratory Landbirds and Estimation of Cumulative Effects of Human Development in the Oil Sands Areas of Alberta