British Columbia Mine Reclamation Symposium

Vegetation quality assessment : measuring quality of vegetation communities to support the accounting… Boyle, B. L.; Gullison, R. E.; Vasiga, D. J.; Luini, G. L.; Franklin, C. W. 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


59367-Boyle_B_et_al_Vegetation_measuring_quality.pdf [ 618.91kB ]
JSON: 59367-1.0374766.json
JSON-LD: 59367-1.0374766-ld.json
RDF/XML (Pretty): 59367-1.0374766-rdf.xml
RDF/JSON: 59367-1.0374766-rdf.json
Turtle: 59367-1.0374766-turtle.txt
N-Triples: 59367-1.0374766-rdf-ntriples.txt
Original Record: 59367-1.0374766-source.json
Full Text

Full Text

VEGETATION QUALITY ASSESSMENT: MEASURING QUALITY OF VEGETATION COMMUNITIES TO SUPPORT THE ACCOUNTING METRICS OF THE BIODIVERSITY VISION OF NET POSITIVE IMPACT FOR A LARGE-SCALE MINING OPERATION   B.L. Boyle, PhD1 R.E. Gullison, PhD.2 D.J. Vasiga, BSc.3 G.L. Luini, BSc.4 C.W. Franklin, MSc.5  1Research Scientist, Dept. of Ecology & Evolutionary Biology, The University of Arizona Tucson, Arizona  2Biodiversity and Ecosystems Specialist, Natural Capital Consulting Inc. Nanaimo, British Columbia  3Senior Lead, Environmental Data Management, Teck Coal Limited Sparwood, British Columbia  4 Senior Coordinator, Environment, Teck Coal Limited Sparwood, British Columbia  5Acting Manager, Environmental Operations, Teck Coal Limited Sparwood, British Columbia   ABSTRACT  Natural resource projects are increasingly governed by requirements that biological impacts be balanced by gains from mitigation measures such as avoidance, minimization, reclamation and biodiversity offsets. Losses and gains are typically quantified in terms of the extent and "quality" or condition of the vegetation component within an ecosystem. Existing measurement frameworks share a number of weaknesses including a reliance on expert opinion, failure to incorporate natural variation, and inability to measure uncertainty or control for bias. We present the Vegetation Quality Assessment (VQA) framework, a new approach that measures quality as the overlap between the sampling distributions of ecological indicators at the project site and those of reference (benchmark) vegetation of the same kind. Stratified-random sampling reduces bias and enables inference for entire classes of vegetation at the level of the landscape. Indicators are commonly-measured attributes such as species richness, taxonomic composition, and percent cover by growth form. When combined, the individual indicator scores provide an index of overall vegetation quality, yielding an intuitive and easily-visualized measure of quality in terms of percent similarity to benchmark. We demonstrate the VQA framework at Teck Coal Limited (Teck) mine operations in the Elk Valley of southeastern BC. This information will provide a critical baseline for determining reclamation targets, designing offsets, and assessing progress towards Teck’s vision of aiming for net positive impact on biodiversity in areas where they operate.  KEYWORDS  Habitat quality, ecosystems, conservation, restoration, plant communities, natural variation   INTRODUCTION  Society and regulatory agencies are increasingly requiring that industrial projects result in no net loss (NNL) or even a net positive impact (NPI) on biodiversity. At the same time, many companies recognize the strong business case to be made for implementing proactive voluntary programs of avoidance, mitigation, restoration and offsets (Rainey et al. 2015). In many programs, the extent and quality of native vegetation is the principal currency by which impacts to terrestrial ecosystems are measured (Gibbons and Freudenberger 2006, Eyre et al. 2011). In addition to their inherent value as a major part of the world's biodiversity, plants supply the energy flows and habitats upon which other organisms depend, and provide numerous environmental services essential to human prosperity (Schlesinger and Andrews 2000, Kier et al. 2005). An accurate, objective and cost-effective mechanism for measuring vegetation quality should thus be a key component of biodiversity management plans that target NNL or NPI.    In many biodiversity assessment frameworks, “vegetation quality" refers to the degree to which vegetation at a site resembles native vegetation in the absence of human disturbance (Parkes and Newell 2003, Gibbons and Freudenberger 2006). Vegetation structured by natural disturbances (such as flooding and avalanches) or maintained by pre-European human intervention (such as fire-managed grasslands in North American and Australia) are also generally considered native vegetation (Gibbons et al. 2008b). Although the importance of native vegetation as a focus of conservation and restoration is widely recognized and largely uncontroversial (Gibbons and Freudenberger 2006, Landres et al. 2014), debate still surrounds the question of how to measure vegetation quality (McCarthy and Parris 2004, Gibbons and Freudenberger 2006, Cook et al. 2010).  As part of Teck Coal Limited’s (Teck) long-term goal of achieving a NPI on biodiversity (Franklin et al. 2018) the company sought to develop an accurate and transparent accounting framework for quantifying impacts and improvements in the extent and quality of native vegetation. Although it would be developed in the specific context of Teck’s Elk Valley coal operations in southeastern British Columbia, the framework needed to be transferable to all areas of operation, both inside Canada and abroad. The focus would be on natural ecosystems, with quality measured empirically relative to a benchmark of undisturbed native vegetation. For maximum accuracy and credibility, the measure of quality would reflect not only average conditions but also the range of natural variation present at the landscape scale. In addition, methods needed to be sufficiently sensitive to distinguish major vegetation types and their successional stages, as well as different classes and intensities of disturbance. Changes in vegetation quality and extent would be measured relative to a baseline of pre-mine conditions. As multiple assessment cycles would be required to document baseline conditions, quantify impacts, monitor recovery and identify the point at which gains exceed losses (and NPI is achieved), field methods needed to be as efficient and cost-effective as possible. Ideally, the framework would be compatible with existing methods, allowing the use of previously collected vegetation data. Most importantly, the framework needed to be statistically sound and scientifically defensible; in other words: (a) repeatable and verifiable, (b) minimally vulnerable to observer bias, (c) capable of estimating uncertainty, and (d) sensitive to changes in quality at temporal and spatial scales relevant to  mining operations, from initiation to closure, and in some cases beyond. In this paper, we describe the development of this system, which we call the Vegetation Quality Assessment (VQA) framework.   BACKGROUND  We identified several existing workflows that track gains and losses in vegetation or habitat quality. These included Habitat Hectares (Parkes and Newell 2003), BioCondition (Eyre et al. 2011), Biometric (Gibbons et al. 2008a) and Quality Hectares (Rio Tinto 2008), among others. A common feature of these methods is the use of a condition or quality to discount the actual area of a given type of vegetation. We follow Rio Tinto (2008) in referring to quality-discounted area as “Quality Hectares” (QH). Quality is typically determined for multiple attributes of the vegetation or the surrounding physical environment, and the individual indicator scores combined to produce a single overall quality score for the vegetation or site being assessed. In all methods, indicator quality is determined relative to an empirical benchmark representing the average value (or range of values) for that indicator in undisturbed vegetation of the same type (Parkes and Newell 2003). Benchmark values are determined either by sampling undisturbed vegetation adjacent to the project site or by reference to standard values for each vegetation or habitat type, as provided by regulatory agencies (The State of Queensland 2014). We found these approaches to be useful, and have incorporated them into the method.  Unfortunately, all methods failed to meet one or more of the requirements described above. The most common shortcoming was lack of a statistical framework. Sample plots were located deliberately rather than at random, and replication, if performed at all, was typically on the order of 2-5 plots per vegetation type (e.g., Eyre et al. 2011, p.12). Such sample sizes are inadequate to support inferences of quality at the landscape scale. Furthermore, low replication combined with a lack of randomization increases vulnerability to observer bias, limits repeatability, and reduces confidence in the results obtained.   Determination of benchmark values and scoring of indicator quality similarly suffered from the lack of a statistical approach. Benchmark values, when provided in advance as regulatory standards, were simple averages or ranges (The State of Queensland 2014). Indicator quality was scored for individual plots on ranked scales of 1-5 categories representing increasing amounts of difference from the supplied benchmarks (Parkes and Newell 2003). Quality scores of the different indicators were combined at the plot level as a “site score”, and overall quality was in most cases the arithmetic average of the site scores. No attempt was made to assess error or assess statistical significance (for example, using a t-test to determine if the mean indicator value differs between a sample of plots from the project site and another sample from undisturbed vegetation). Overall, the approaches examined provide no basis for determining the degree to which the small number of plots sampled represented the average condition of the entire project site; nor can we be certain to what extent the vegetation of the project site as a whole differs from undisturbed vegetation.  A second shortcoming of existing approaches was the use of field methods and indicator measurements specific to each approach. In some approaches, indicators sampled are selected based expert opinion, either of the author of the method (Parkes and Newell 2003) or of multiple authorities (Oliver et al. 2007). In addition, taxonomic identifications are commonly omitted to enable surveys to be "undertaken rapidly by a range of natural resource managers…not just botanical ecologists" (Parkes and Newell 2003). In theory, these field protocols are designed to be faster than traditional vegetation inventory methods. In our experience, however, the estimated time to complete these specialized surveys (e.g., Eddy et al. 2011) differs little from the time required for standard vegetation surveys such as the Site Visit methodology used extensively in British Columbia (BC Ministry of the Environment 2010). As we had access to a large database of Site Visit inventories collected previously, we preferred a method that would be compatible with these data. Furthermore, we hoped to leverage as benchmark data the large database of inventories maintained by the British Columbia Ministry of Forests, Lands, Natural Resource Operations and Rural Development (BC FLNRORD) and used to develop the BC provincial Biogeoclimatic Ecosystem Classification (BEC). Finally, as there is abundant evidence demonstrating that indicators requiring taxonomic identifications, such as species composition and relative abundance of individual species, are highly sensitive indicators of forest health, successional status and disturbance (Gentry 1988, Kappelle 1995, Letcher and Chazdon 2009), and given that this information is routinely collected as part of the Site Visit protocol, we saw no advantage to omitting taxonomic identifications. For the reasons outlined above, we preferred a framework that would be compatible with existing, commonly used vegetation sampling methods, including the Site Visit methodology.  METHODS  We developed an approach, the VQA framework, which combines useful features of existing approaches—specifically, quality-discounted area, compound quality scores based on individual indicator scores, and empirical benchmarking against measured attributes of native vegetation—with a stratified-random sampling design that improves accuracy, enables measurement of error, and reduces observer bias. Each unique class of vegetation plus disturbance is treated as a sampling stratum. Indicator values for the project site are measured by samples of plots located at random within the strata. Benchmark indicator values are measured by samples of plots from undisturbed examples of the same vegetation present at the project site. Quality of each indicator within each sampling stratum is measured in terms of similarity to the benchmark condition for that indicator. As described below, we use the overlap between the project site and benchmark sampling distributions as the principal measure of similarity, a novel approach that incorporates natural variation at the landscape scale.  Overview of the framework: baseline condition, losses, gains, and progress to NPI  The VQA framework consists of successive assessments of area x quality (QH) of all classes of vegetation and disturbance present at the project site. At a minimum, at least two assessments are required: a baseline assessment of vegetation QH prior to all project impacts, and a final assessment demonstrating that the QH of each vegetation type equals (NNL) or exceeds (NPI) the baseline QH. In practice, as different types of vegetation recover or are restored according to different timelines, multiple assessments are required. If NPI cannot be attained at the project site alone then gains from offset sites may be needed to substitute for residual losses of extent and/or quality at the project site. Measuring such gains requires a separate set of assessments for each offset site.   In general, restoration seeks to reestablish vegetation of the same type that existed prior to project impacts. However, a different target vegetation may be appropriate if a project alters the topography and soils to the extent that the baseline vegetation types can no longer grow. If a different restoration target is chosen, there is generally an expectation that the replacement vegetation is of equal or greater conservation or cultural value than the original vegetation ("like-for-like or better"; Quétier and Lavorel 2011).  For example, a previously forested site converted to mine tailings or waste may be restored to native grasslands in regions where grasslands are of greater conservation concern or are an important resource for valued wildlife species.  Mapping and classification of vegetation and disturbance  Each VQA assessment cycle requires a comprehensive map of all classes of vegetation at the project site. Forested vegetation may be further subdivided into naturally occurring successional classes (seral stages). Each type of vegetation (plus seral stage) is further subdivided into any measurably distinct disturbance classes observed. Each class of vegetation plus disturbance represent a sampling stratum, within which the appropriate number of sampling localities (plots) are distributed at random using the appropriate sample size (See Error, sensitivity and sampling requirements, below). We refer to these plots from the project site as “focal plots”. Similar numbers of plots are also required to represent undisturbed examples of each class of native vegetation present at the project site; we refer to these as “benchmark plots”. Benchmark-quality vegetation may be sampled in areas surrounding the project site, or they may be drawn from existing data repositories such as the BCFLNRORD BEC vegetation database.  Ideally, a baseline map of vegetation and disturbance is completed prior to initiation of any project impacts. However, in cases where a VQA is implemented after the project has begun, modeling (“back-casting”) may be used to reconstruct the vegetation present prior to project start. As the VQA framework was developed in the context of Teck’s Elk Valley operations—where extensive mining activity had already taken place—we used a back-casting approach to reconstruct prior to initiation of mining-related impacts. See Knopff and Franklin (2018) for a detailed description of back-casting methods and construction of baseline vegetation.  All vegetation units were initially classified to the Site Series level, using the BEC version 10 (British Columbia Ministry of Forests, Lands and Natural Resource Operations 2016). In addition, forested vegetation types for both current and back-cast vegetation were assigned structural stage classes using BC provincial standard terminology for describing terrestrial ecosystems (BC Ministry of the Environment 2010). As we did not have a priori information on vegetation condition at baseline year, and historical data (e.g., aerial photos) did not provide clearly interpretable evidence of disturbance, we did not attempt to split vegetation units further into classes of disturbance.   Plot methodology and indicators  As a large number of pre-existing vegetation inventories had already been completed within the boundaries of Teck’s Elk Valley operations (“study area”), we used these pre-existing data for the initial VQA application. The final database consisted of over 1,129 focal plots conforming to the standard Site Visit methodology (BC Ministry of the Environment 2010). In addition, 884 benchmark plots representing undisturbed examples of vegetation from the study area were obtained from the BC FLNRORD BEC vegetation database.   From the dozens of measurements recorded among the nearly 2,000 vegetation plots, we selected 10 indicators (Table 1) that have been shown in the ecological literature to be sensitive indicators of vegetation successional status and disturbance. The indicators are a mixture of taxonomic attributes (species richness, taxonomic composition, percent cover of exotic species), vegetation structure (percent cover by different growth forms) and aspects of the physical environment relevant to ecosystem function (percent ground cover of dead wood, organic soil and surface water). Indicators were grouped into functional categories reflecting these relationships. As described below, functional categories play an important role in combining individual indicator quality scores into an index of overall vegetation quality. As the indicators selected are commonly recorded in vegetation plots, the VQA framework should be applicable to a wide variety of sampling methodologies. Going forward, new plots sampled solely for the purpose of VQA can be completed more efficiently by sampling only these 10 indicators.   Table1. Ecological indicators used for the VQA application.  Indicator Functional category Native species richness Composition Taxonomic composition Composition Percent cover exotic species Composition Percent cover herbs Structure Percent cover moss Structure Percent cover shrubs Structure Percent cover trees Structure Percent cover dead wood Function Percent cover organic soil Function Percent cover surface water (wetland vegetation only) Function   Indicator quality  For each indicator within each land cover class, we used the overlap between the focal and benchmark sampling distributions as the measure of quality (Figure 1). This was done by fitting separate probability distributions to the focal and benchmark data, using the appropriate distribution. For example, species richness was fit using a negative binomial distribution; measures using percent cover were converted to proportions and fit using the beta distribution. Quality, calculated as the overlap between the two curves, varies between 0 and 1 (0 to 100%) and provides a direct and intuitive measure of the degree to which both overlap and variance of the focal distribution resemble that of the benchmark distribution for the indicator in question. Comparing distributions rather than means incorporate natural variation at the landscape scale and allows non-normally-distributed indicators to be compared directly, without the need for transformations or assumptions of normality.   Overall vegetation quality  Overall vegetation quality was calculated in two stages. First, we calculated the arithmetic mean of the indicator quality scores within each functional category. Second, overall quality was calculated as the geometric mean of the functional category means. Because the geometric mean drops rapidly to zero as one or more terms approach zero, this precludes quality in one functional category from completely substituting for low to zero quality in another category. For example, using only arithmetic means, a reclaimed site with high quality scores for organic soil and dead wood but no living plants would still have a non-zero quality score. However, using a geometric mean overall quality of the same site would be zero. This approach addresses criticisms of substitutability that have been directed at other approaches such as Quality Hectares (McCarthy & Parris 2004).  Estimation of baseline quality  Focal plots from the Elk Valley study area were sampled over a 41-year period post-dating the baseline year, from 1975-2016. All plots were sampled in vegetation unaffected by mining activity. Timber extraction and other related forms of anthropogenic disturbance ended at or were prior to the baseline year (when Teck’s stewardship began). We therefore used the quality Figure 1. Overlap between two indicator sampling distributions as a measure of quality. The distribution of the indicator being compared, species richness, is a continuous approximation of the negative binomial distribution. (a) High overlap, high quality; (b) Low overlap, low quality. measured in these plots as proxies for quality at baseline year. As quality at these sampling locations most likely would have improved since baseline year due to natural regeneration, the effect of assuming higher quality at baseline year would be to over-estimate quality at baseline year. We believe this assumption is justified as it is conservative (i.e., provides a high estimate) with respect to Teck’s impacts and resulting mitigation obligations.  Error, sensitivity and sampling requirements  We used the bootstrap 95% confidence intervals of indicator and overall quality as measures of error and sensitivity, using an empirical bootstrap approach (Effron 1979). The difference between the mean and confidence limits (CL), or margin of error, represents the smallest difference that can be detected at a given sampling intensity and provides an intuitive index of empirical effect size. As confidence limits were not symmetrical about the mean for the majority of indicators, the largest of the two margins of error was used as the minimum detectable effect size. For the current application, we selected an effect size of 0.1 (i.e., the ability to detect a change in quality of at least of 10%) as the target level of sensitivity. Minimum sample sizes requirements were determined using a Monte Carlo simulation approach (Effron 1979) to examine the relationship between sample size and power for empirical effect sizes from 0.1 to 0.2 (i.e., 10% to 20% changes in quality, respectively), resampling the actual data over a range of sample sizes. For a given sample size, we performed a simulation of 1000 random draws of the actual data, calculating the empirical effect size (as described above) for each draw. Power was determined as the proportion of draws out of 1000 with an observed effect size ≤ the target effect size.  A power-sample size curve was then constructed by fitting a logistic curve to the full set of simulations across all sample sizes. The minimum required sample size (nmin) was then estimated as the smallest sample size that surpassed a conventional power of 0.80 at a given effect size. We performed separate calculations of nmin for overall quality and for each indicator in each vegetation class. These estimates of nmin will be used to determine the numbers of additional samples needed to complete the baseline assessment, and to set sampling targets for later assessment cycles as we monitor recovery and progress to NPI.  Software  All analyses were performed in R (R Development Core Team. 2008). Probability distributions were fit using a maximum likelihood approximation, as implemented in R package MASS (Ripley et al. 2013).  RESULTS & DISCUSSION  Revisions to vegetation classification and baseline map  Completion of the initial baseline vegetation map resulted in 301 vegetation types (site series) within the project boundaries. As most plots matched to only a small number of site series, this left many site series unsampled or represented by very low sample sizes. We therefore reduced the number of sampling strata by combining site series into larger “ecosystem groupings” representing vegetation of similar composition and structure (Table 2). For example, all site series representing primarily shrubby alpine vegetation was combined into the ecosystem groupings “Alpine dwarf shrub”. We also reduced the number of forest seral stage classes to three: “Early-mid”, “Mature” and “Old growth”.   Table 2. Example of three ecosystem groupings showing their component site series. Ecosystem groupings rather than site series were used as the classification for the final baseline map to reduce the number of sampling strata and increase plot sample sizes per stratum.  Site series Ecosystem grouping ESSFdk1/Af Alpine ESSFdk1/Ag Alpine ESSFdk1/Am Alpine ESSFdkw/Ag Alpine ESSFdkw/Ah Alpine ESSFdkw/Am Alpine ESSFdkw/At Alpine ESSFdmw/Ag Alpine ESSFdkp/Af Alpine dwarf shrub ESSFdkp/Ah Alpine dwarf shrub ESSFdkp/At Alpine dwarf shrub ESSFdkw/Af Alpine dwarf shrub ESSFwmp/Ah Alpine dwarf shrub IMAun/Af Alpine dwarf shrub IMAun/Ah Alpine dwarf shrub IMAun/At Alpine dwarf shrub ESSFdk2/Ag Alpine grassland ESSFdkp/Ag Alpine grassland ESSFwm1/Ag Alpine grassland ESSFwmw/Ag Alpine grassland IMAun/Ag Alpine grassland  Converting site series to ecosystem groupings and separating forested ecosystem groupings into the three seral stages resulted in a total of 21 distinct sampling strata (vegetation plus seral stage) for the final baseline map. Revised sample sizes were robust (most >20 plots, and many much larger) for all but four strata (Table 3).  Table 3. Vegetation (ecosystem groupings) and seral stages of final baseline map sampling strata, showing the number of focal and benchmark plots in each stratum.   Ecosystem grouping Focal plots Benchmark plots Alpine 82 6 Alpine dwarf shrub 30 46 Alpine grassland 41 4 Ecosystem grouping Focal plots Benchmark plots Alpine meadow 15 2 Avalanche feature 24 69 Brushland/Grassland 61 60 Deciduous floodplain 4 0 Dry forest, early-mid 95 76 Dry forest, mature 70 53 Dry forest, old 4 13 Herb meadow 6 0 Intermediate forest, early-mid 225 140 Intermediate forest, mature 216 77 Intermediate forest, old 22 27 Krummholz 9 15 Rock/Talus 61 58 Shrubland 6 2 Wet forest, early-mid 46 66 Wet forest, mature 54 73 Wet forest, old 12 27 Wetland 46 70   Overall vegetation quality  Baseline overall quality (95% confidence limits in parentheses) for vegetation within the Elk Valley study area (Table 4) ranged from 77.5% (67.0-86.7%) for old growth dry forest to 92.4% (87.3-94.9%) for avalanche vegetation (“Avalanche feature”). In general, quality scores were lowest for forested vegetation. This difference between the Elk Valley forests and undisturbed forest is consistent with a history of forest extraction prior to Teck’s tenure, and is further supported by examination of the quality scores of individual indicators (see Indicator quality, below).  Although the majority of quality scores were relatively high, no vegetation type had upper confidence limits that overlapped 100%; thus, even apparently undisturbed upland vegetation such as Alpine and Alpine Dwarf Shrub differed significantly, if slightly, from benchmark. Whether these differences are entirely due to human disturbance or at least in part to natural, locally-distinctive traits of native vegetation will require careful examination of the taxonomic composition of individual plots. However, the generally low values of quality for taxonomic composition relative to other indicators (74.8 ± 18.38% SD) lends some weight the latter explanation and suggest that additional filtering of benchmark data to include only samples from the local region may be warranted.  Quality hectares  Table 4 also provides a comparison between actual area and QH for the baseline year. Total area for the entire study area, excluding anthropogenic/non-vegetated sites is 18,978.9 ha.  Total QH for vegetation types that had sufficient samples to be assessed was 16,697.9 QH.  Using lower confidence limits for the assessed vegetation classes and assuming a quality of 0% for the remaining non-assessed vegetation gives a low estimate of 16,053.7 QH. Using upper 95% confidence limits for the assessed vegetation and assuming a quality of 100% for the non-assessed vegetation gives a high estimate of 17,484.6 QH.  However, it should be stressed that different vegetation types are generally not exchangeable, and total QH is therefore less meaningful as a mitigation target than the baseline QH of the individual vegetation types. In other words, achieving NPI means ensuring that where possible QH of each vegetation type match or exceed the baseline QH of that same vegetation type, as shown in Table 4. As discussed earlier, the concept of “like for like, or better” may be applied when certain vegetation types are difficult or impossible to mitigate for based on changes from disturbance to topography, hydrological cycles, etc. In this context, the strategy for ecosystem mitigation in the Elk Valley may look at what vegetation types are best suited at a landscape level. This approach requires careful consideration and is discussed further in Franklin et al., (2018).  Indicator quality  Examination of raw scores and quality for individual indicators provides insight into overall quality score of different vegetation types. For example, early- to mid-seral wet forest had lower species richness and differed significantly in taxonomic composition, compared to natural early- to mid-seral wet forest (Figure 2). Furthermore, as shown by the sampling distribution for indicator “Percent cover trees”, a high proportion (nearly 40%) of the focal sites within the Elk Valley had very low tree cover. These results are consistent with a history of logging at the site. In addition, the bimodal distribution for “Percent cover trees”, with strong inflation near zero, indicates that this vegetation type is a mix of two different disturbance histories. If possible, it may be desirable to split this class into two classes, possibly representing logged and unlogged forest.  The vegetation type “Wetland” provided another example of the information to be gleaned from inspecting individual indicators. Wetland was the only vegetation type for which species richness of the study area vegetation exceeded that of the benchmark vegetation (Figure 3). However, despite greater species richness, taxonomic composition could not be distinguished from benchmark (Taxonomic distance quality confidence limits 89.7-109.2%). In addition, focal plots in wetland vegetation were on average drier than benchmark plots, with nearly twice as many focal plots having a percent cover of surface water of 25% or less. Taken together, these results are consistent with the explanation that some focal plots may be mixtures of different vegetation types represented by separate plots in the benchmark data. Another possibility is that the focal plots include drier and/or more disturbed wetland vegetation. In either case, the taxonomic composition of the individual plots should be inspected carefully to determine if some are misclassified, or if the entire “Wetland” vegetation type should be split into multiple, more homogeneous vegetation types.Table 4. Baseline quality and quality hectares of vegetation (ecosystem groupings) within the Elk Valley study area. CLs: confidence limits; ES: effect size;   Land cover class Quality 95% CLs ES Actual area (ha) QH 95% CLs Notes Anthropogenic/non-vegetated    534.7   Not included in totals Native vegetation (ecosystem grouping)        Alpine 91.1 85.8-96.8 0.057 385.0 350.8 330.2-372.9   Alpine dwarf shrub 87.0 82.2-91.0 0.048 0.6 0.6 0.5-0.6   Alpine grassland    67.1   Insufficient data  Alpine meadow    13.0   Insufficient data  Avalanche feature 92.4 87.3-94.9 0.050 741.5 685.1 647.7-703.6   Brushland/Grassland 90.5 85.9-94.9 0.046 862.0 780.1 740.8-817.8   Deciduous floodplain    157.3   Insufficient data  Dry forest, early-mid 91.1 88.2-93.6 0.028 920.4 838.1 812.1-861.7   Dry forest, mature 89.3 86.1-92.6 0.033 521.4 465.7 449.2-482.7   Dry forest, old 77.5 67.0-86.7 0.105 94.1 72.9 63.1-81.6   Herb meadow    0.3   Insufficient data  Intermediate forest, early-mid 88.6 86.6-90.6 0.020 7771.0 6889 6730.7-7041.2   Intermediate forest, mature 91.8 87.4-94.8 0.044 3931.0 3611 3436.1-3727.1   Intermediate forest, old 85.4 81.3-90.3 0.049 1272 1086 1034.1-1148.1   Krummholz 86.4 78.8-92.0 0.076 55.6 48 43.8-51.1   Rock/Talus 89.3 86.3-91.6 0.030 233.4 208.4 201.4-213.7   Shrubland    9.4   Insufficient data  Wet forest, early-mid 82.0 76.9-85.3 0.051 923.1 757.2 709.8-787.7   Wet forest, mature 89.2 85.0-92.7 0.042 565.0 504.2 480.5-523.7   Wet forest, old 87.8 80.5-93.3 0.073 317.0 278.2 255.2-295.7   Wetland 88.8 85.4-92.5 0.037 138.8 123.2 118.5-128.3  Sensitivity, effect size and sample size requirements  Uncertainty of the estimates of overall quality, as measured by the 95% confidence intervals, was highest for old growth dry forest (19.7%) and lowest for early-mid intermediate forest (4.0%), with effect sizes ranging from 0.02 for early-mid intermediate forest to 0.11 for old growth dry forest (Table 4). Old growth dry forest also had the lowest focal sample size among all assessed vegetation types, with only four focal plots (Table 3). With the exception of the latter vegetation, effect sizes for overall quality were all <0.10, indicating that available sample sizes met or exceeded the target sensitivity with respect to overall quality. Simulations of sample size and power for overall quality confirmed these relatively modest sampling requirements. A maximum sample size of 10 (that is, 10 focal and 10 benchmark plots) was sufficient to detect a 10% change in overall quality for all vegetation types (Figure 5).    Observed confidence intervals for indicator quality were much more variable, ranging from effectively zero for “Percent cover exotic species” in vegetation types with no exotic species (e.g., Krummholz, old growth intermediate and dry forest) to 20% or more in vegetation with low sample sizes (e.g., 40.7% for “Species richness” in “Dry forest, old”). Minimum sample sizes, as determined by the simulations, were also highly variable. For Krummholz and Old Growth Intermediate Forest, “Percent cover exotic species” was zero in all focal or benchmark plots, and and power of 0.80 was exceeded at all simulated sample sizes. By comparison, the results shown in Figure 6 for Species Richness” in “Old Growth Wet Forest” are representative of the most variable vegetation-indicator combinations. Sample sizes of 80 plots or more were required to achieve a power ≥0.80 at the target effect size of 0.1. In general, however, a target empirical effect size of 0.15 could be achieved for nearly all vegetation-indicator combinations with sample sizes of 35-40 plots.    CONCLUSIONS  The trial application at Teck operations in the Elk Valley was the first baseline application of what will be multiple assessments of vegetation quality as we measure mining impacts and monitor progress to NPI. As residual impacts from the mine itself will likely be considerable, we will also likely need to apply the same approach to offset sites. Nonetheless, based on this first iteration, we are confident that the VQA framework can provide a repeatable, transparent and objective accounting system for measuring gains andlosses in vegetation quality and extent. Despite the modest sample sizes required to measure overall quality (10 focal and 10 benchmark plots) we recommend more intensive sampling to allow measurement of indicator quality with adequate power and sensitivity. Although a target effect size of 0.10 for indicator quality may be an unreasonable goal given the high sampling requirements for some indicator-vegetation combinations (≥80 plots) a target effect size of 0.15, requiring roughly 35-40 plots, should be more   Figure 3. Focal and benchmark distributions of (a) “Percent cover water” and (b) “Species Richness” for vegetation “Wetland”.    readily attainable. As this analysis has shown, the individual indicator distributions and their quality scores provide important insights as to why a particular vegetation type differs from the benchmark condition. This information can be used to guide reclamation activities to speed succession and recovery; for example, through targeted removal of exotic species, or clearing of shrubby vegetation encroaching on grasslands.  Fortunately, the small number of indicators required for a VQA plot means that, going forward, new plots can be installed relatively rapidly. Furthermore, as reference databases grow, little effort will be devoted to sampling new benchmark plots. The latter data can be re-used for different projects at different locations, as long as the native vegetation types being assessed are the same.  Figure 5. Monte Carlo simulations of sample size and power for indicator “Species Richness” in “Old Growth Wet Forest”. Solid lines are logistic fits to simulations at each of three empirical effect sizes. The minimum sample size (n.min) needed to achieve a conventional power of 0.80 is indicated by the vertical line.    ACKNOWLEGEMENTS  We thank Alison Burton and Justin Straker for providing insightful suggestions that helped guide development of the VQA framework.   REFERENCES   BC Ministry of the Environment. 2010. Field Manual for Describing Terrestrial Ecosystems. 2nd Editio. B.C. Ministry of Forests and Range, B.C. Ministry of Environment. British Columbia Ministry of Forests-Lands and Natural Resource Operations. 2016. Biogeoclimatic Ecosystem Classification, Version 10. Cook, C. N., G. Wardell-Johnson, M. Keatley, S. a. Gowans, M. S. Gibson, M. E. Westbrooke, Figure 4. Monte Carlo simulations of sample size (number each of focal and benchmark plots) and power to detect a change in overall vegetation quality equivalent to an empirical effect size of 0.1. Each point is an estimate of overall quality for one vegetation type, based on a random draw of the actual data at the sample size shown. Lines join points of the same vegetation type. Vertical dashed lines bracket the range of minimum sample sizes (n.min) where the simulation lines cross the horizontal line marking the conventional power of 0.80.  and D. J. Marshall. 2010. Is what you see what you get? Visual vs. measured assessments of vegetation condition. Journal of Applied Ecology 47:650–661. Eddy, D., R. Hall, R. Rehwinkel, G. Baines, J. Dorrough, M. Austin, A. L. Kelly, A. J. Franks, and T. J. Eyre. 2011. Assessing the assessors : Quantifying observer variation in vegetation and habitat assessment 12:144–148. Efron, B. 1979. Bootstrap Methods: Another Look at the Jackknife. The Annals of Statistics 7:1–26. Eyre, T. J., A. . Kelly, V. J. Neldner, B. A. Wilson, D. J. Ferguson, M. J. Laidlaw, and A. J. Franks. 2011. BioCondition: A Condition Assessment Framework for Terrestrial Biodiversity in Queensland. Assessment Manual. Version 2.1. Page Environment. Brisbane. Franklin, C. W., S. R. Hilts, and R. E. Gullison. 2017. Teck’s recent experience in pursuing net positive impact (NPI) for biodiversity at coal mines in BC and Alberta. Teck Coal Limited, Sparwood, B.C. Gentry, A. H. 1988. Changes in plant community diversity and floristic composition on environmental and geographical gradients. Annals of the Missouri Botanical Garden 75255:1–34. Gibbons, P., D. Ayers, J. Seddon, S. Doyle, and S. Briggs. 2008a. Biometric 2.0: A Terrestrial Biodiversity Assessment Tool for the NSW Native Vegetation Assessment Tool - Operational Manual. Page Assessment. Canberra. Gibbons, P., S. V. Briggs, D. a. Ayers, S. Doyle, J. Seddon, C. McElhinny, N. Jones, R. Sims, and J. S. Doody. 2008b. Rapidly quantifying reference conditions in modified landscapes. Biological Conservation 141:2483–2493. Gibbons, P., and D. Freudenberger. 2006. An overview of methods used to assess vegetation condition at the scale of the site. Ecological Management and Restoration 7:S10–S17. Kappelle, M. 1995. Ecology of mature and recovering Talamancan montane Quercus forests, Costa Rica. University of Amsterdam, AmsterdamEditor. Kier, G., E. Dinerstein, T. H. Ricketts, W. Küper, H. Kreft, and J. M. Wilhelm Barthlott. 2005. Global patterns of plant diversity and floristic knowledge. Journal of Biogeography 32:1107–1116. Knopff, K., and W. E. Franklin. 2017. Biodiversity Management: Establishing Pre-Existing Baseline Conditions on Mature and Historical Mining Disturbances to Derive Back-casted Wildlife Habitat Suitability Model Metrics for Ten Wildlife Species. Teck Coal LImited, Sparwood, BC. Letcher, S. G., and R. L. Chazdon. 2009. Rapid Recovery of Biomass, Species Richness, and Species Composition in a Forest Chronosequence in Northeastern Costa Rica. Biotropica 41:608–617. McCarthy, M., and K. Parris. 2004. The habitat hectares approach to vegetation assessment: An evaluation and suggestions for improvement. Ecological Management & Restoration 5:24–27. McElhinny, C., P. Gibbons, C. Brack, and J. Bauhus. 2005. Forest and woodland stand structural complexity: Its definition and measurement. Forest Ecology and Management 218:1–24. Oliver, I., H. Jones, and D. L. Schmoldt. 2007. Expert panel assessment of attributes for natural variability benchmarks for biodiversity. Austral Ecology 32:453–475. Parkes, D., and G. Newell. 2003. Assessing the quality of native vegetation: the “habitat hectares” approach. Ecological Management & Restoration 4:29–38. Quétier, F., and S. Lavorel. 2011. Assessing ecological equivalence in biodiversity offset schemes: Key issues and solutions. Biological Conservation 144:2991–2999. R Development Core Team. 2008. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Rainey, H. J., E. H. B. Pollard, G. Dutson, J. M. M. Ekstrom, S. R. Livingstone, H. J. Temple, and J. D. Pilgrim. 2015. A review of corporate goals of No Net Loss and Net Positive Impact on biodiversity. Oryx 49:232–238. Rio Tinto. 2008. Rio Tinto and biodiversity: achieving results on the ground. Rio Tinto plc and Rio Tinto Limited. Ripley, B., B. Venables, D. M. Bates, K. Hornik, A. Gebhardt, and D. Firth. 2013. Package ‘MASS.’ Cran R. Schlesinger, W., and J. Andrews. 2000. Soil respiration and the global carbon cycle. Biogeochemistry 48(1):7–20. The State of Queensland. 2014. BioCondition Benchmarks.   


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items