Woodland caribou scientific review to identify critical habitat: chapter 15

Appendix 6.5 (Continue)

A National Meta-analysis of Boreal Caribou Demography and Range Disturbance

Results

Estimates of population condition

Recruitment was positively correlated with population rate of change for both the subset of data evaluated here (r=0.75; p<0.01) and the Sorensen et al. (2008) data (r=0.63; p<0.01). Regression analysis yielded very similar constants and coeffi cients (Table 2). Recruitment was not correlated with adult female survival in either data set. Exploratory analysis of the subset of 15 populations further revealed recruitment to be more sensitive to % anthropogenic disturbance and % total disturbance than either adult female survival or population growth rate. Use of recruitment as an index of population condition for subsequent analyses of main models therefore seems reasonable.

Appendix 6.5 - Table 2. Regression statistics for analysis of mean annual recruitment versus population growth rate for a 15 population subset of data compiled for this study and 6 Alberta populations included in Sorensen et al (2008).

Data Source R2

β0

intercept

SE P

β 1

(X1 )

SE P
15 population subject 0.56 0.84 0.030 <0.001 0.005 0.001 0.001
Sorenson et al. (2008) 0.40 0.84 0.033 <0.001 0.007 0.002 <0.001

Regression diagnostics and data selection for main models

For the full data set, outliers were examined visually and tested for leverage and infl uence (leverage versus normalized residual squared plots) with DFBETA (STATA 8.0), which assesses how the coeffi cient is affected by deleting each of the observation values (values exceeding 2/sqrt n = 0.4 are of concern). Only Charlevoix had a DFBETA value above the model cut-off in Model 3 (Charlevoix DFBETA = 0.70). Given that was the only data point that signifi cantly affected estimation of the regression coeffi cient, and that it was also the sole reintroduced population, it was excluded from further analyses.

There was no evidence of heteroscedasticity in the residuals of any of the models (White’s test and Breusch-Pagan test, STATA 8.0). Residuals from Models 1 and 2 met conditions of normality; however, residuals from Model 3 signifi cantly deviated from normality (Shapiro-Wilk test of normality, P = 0.01). Log transformations of the variable total disturbance were considered, as well as the addition of a squared term, to examine potential non-linear forms of the relationship. Neither of these options increased the fi t of the model. Therefore, the linear form was retained due to ease of interpretation, and a lack of knowledge concerning the true form of the underlying distribution.

Regression results

There was no signifi cant relationship between caribou recruitment rate and the percent area disturbed by fi re alone (F 1,22 = 2.52, p = 0.13; Model 1, Table 3; Figure 3). However, there were signifi cant negative relationships between recruitment and the percent area affected by anthropogenic disturbance (F1,22= 20.21, p < 0.001; Model 2, Table 3; Figure 4) and with the merged measure of total disturbance (F1,22= 34.59, p <0.001; Model 3, Table 3; Figure 5). Model 3, the measure of total disturbance, had the lowest AICc value and best fi t with population recruitment rates (Table 3, Figure 5).

Appendix 6.5 - Table 3. Regression statistics for analysis of mean annual recruitment versus parameters of range disturbance for boreal caribou populations across Canada (n=24).

Model R2

β0

intercept

SE P

β1

(X 1 )

SE P AICc
1 - % fi re 0.10 31.86 4.10 <0.001 -0.31 0.20 0.13 54.81
2 - % anthropogenic 0.49 39.13 3.40 <0.001 -0.43 0.10 <0.001 49.15
3 - % total disturbance 0.61 46.37 3.75 <0.001 -0.49 0.08 <0.001 46.09

There was no clear pattern between the size of population ranges or study areas and the observed relationship between recruitment and total range disturbance (Figure 6).

Appendix 6.5 - Figure 3. Linear regression of mean caribou recruitment versus the percent of range disturbed by fi re within 50 years of the most recent demographic data (n = 24). The relationship is not signifi cant (P=0.13)

Appendix 6.5 - Figure 3. Linear regression of mean caribou recruitment versus the percent of range disturbed by fi re within 50 years of the most recent demographic data (n = 24). The relationship is not signifi cant (P=0.13)

Appendix 6.5 - Figure 4. Linear regression of mean caribou recruitment versus the percent of range affected by anthropogenic disturbance (n = 24).

Appendix 6.5 - Figure 4. Linear regression of mean caribou recruitment versus the percent of range affected by anthropogenic disturbance (n = 24).

Appendix 6.5 - Figure 5. Linear regression of mean caribou recruitment versus the percent of range affected by fi re and anthropogenic disturbance, accounting for spatial overlap of the variables (n = 24).

Appendix 6.5 - Figure 5. Linear regression of mean caribou recruitment versus the percent of range affected by fi re and anthropogenic disturbance, accounting for spatial overlap of the variables (n = 24).

Appendix 6.5 - Figure 6. Linear regression of mean caribou recruitment versus the percent of range disturbed by fi re and anthropogenic disturbances, accounting for spatial overlap of the variables (n=24). The size of circles represents the relative size of individual ranges or study areas (see Table 1).

Appendix 6.5 - Figure 6. Linear regression of mean caribou recruitment versus the percent of range disturbed by fi re and anthropogenic disturbances, accounting for spatial overlap of the variables (n=24). The size of circles represents the relative size of individual ranges or study areas (see Table 1).

 

Discussion

This is the fi rst analysis of caribou demography and range disturbance at the scale of the national distribution of boreal woodland caribou in Canada. We found a clear negative relationship between caribou recruitment, as measured by calf/cow ratios, and the level of disturbance within caribou ranges. Total disturbance (non-overlapping burn and anthropogenic disturbance) was the best predictor of boreal caribou recruitment rates. As in Sorensen et al. (2008), the extent of anthropogenic disturbance appeared to be the main driver of this relationship, also refl ecting results from other studies where the level of anthropogenic disturbance infl uenced caribou distribution and persistence (Courtois et al. 2007, Schaefer and Mahoney 2007, Vors et al. 2007, Wittmer et al. 2007).

The relationship between recruitment rate and proportion of range disturbed by fi re was less clear. The percent area burned within caribou ranges was not a signifi cant predictor of recruitment rate by itself, but its merger with the anthropogenic disturbance layer did improve model fi t. Similar to anthropogenic disturbances, fi res affect the amount, composition and age structure of forest available to caribou, although the effect on confi guration may be different; that is, disturbance by fi re tends to be more aggregated and thus result in less fragmentation of remaining areas (e.g., Schmiegelow et al. 2004). Spatially, fi res are represented as polygons of disturbance without consideration of severity; however, fi res in boreal forests are highly variable, and often result in mosaics of burned and unburned patches within the mapped fi re boundary (Smyth et al. 2005, Schmiegelow et al. 2006). This variability is likely to result in differential effects on habitat quality for caribou, dependent on their immediate effects on lichen and other forage, the post-disturbance trajectory of burned areas, and the indirect effects of disturbance by fi re on habitat suitability and resultant numerical response by predators and apparent competitors. Nevertheless, the main question is how disturbance by fi re differs from anthropogenic disturbances with respect to demographic response by caribou. In this regard, a conspicuous difference is the absence of linear features in naturally disturbed areas. As a result, fi res are unlikely to elicit the functional response by predators attributed to increased travel and hunting effi ciency in association with linear anthropogenic disturbances (James and Stuart-Smith 2000, James et al. 2004, Dyer et al. 2001, McLoughlin et al. 2003). They are many other aspects that could be examined, such as post-disturbance successional trajectories following fi re or harvest, but comprehensive treatment is beyond the scope of the present exercise.

One methodological consideration is the 50-year window for quantifying disturbance by fi re. The 50-year interval is consistent with Sorensen et al. (2008), and with the anticipated duration of effects on caribou from several studies (Klein 1982, Schaefer and Pruitt 1991, Dunford et al. 2006), but represents a discrete cut-off when extracting the disturbance data. For example, a large fi re that burned 51 years before the last year for which demographic data were available would not have been included in the disturbance estimate for that local population range. Similarly, 49 year-old and 1 year-old fi res were considered identical within a range, and no consideration was afforded across ranges to potential variability in the duration of impacts. Future analyses should consider a variable or moving window for measuring this disturbance at the level of individual ranges, and given the large geographic extent over which the species is distributed, where possible incorporate information on variability in postfi re regeneration and recovery.

Measures of both anthropogenic and natural disturbance in this study were arguably conservative, due to a requirement to use nationally-standardized data sets. The Global Forest Watch Canada data were restricted to detection of features readily identifi ed from mid-resolution satellite imagery (1:40,000–1:50,000 scale; overall pixel resolution of 28.5 m), and the Canadian Large Fire Database includes only fi res >200 ha in size. Thus, narrow and small disturbances were not captured. Furthermore, the most recent anthropogenic disturbance data included were to 2005, and some features were current only to 2003. Effort was made to match demographic data to the disturbance layers; however, data availability was a constraint. In ranges experiencing high rates of change, the level of disturbance may have been underestimated, particularly when demographic data were very recent. Regardless, the strength of our analyses includes the standardization of disturbance measures across ranges, and the repeatability of the procedure. Finally, while our analyses revealed some fundamental relationships with a parsimonious explanation, our disturbance measures captured only a subset of the attributes that affect range condition, and a better understanding of additional range attributes could help explain variation in the observed relationships at a national scale. It is also important to note that our measures of disturbance accounted only for conspicuous changes to forest cover that could be derived from national-scale data and mapped. Some caribou ranges in Canada experience other forms of disturbance that may compromise population condition and/or affect range use. For example, low level aircraft traffi c can affect caribou reproduction (Luick et al. 1996, Maier et al. 1998) and calf survival (Harrington and Veitch 1992). Over-hunting can also drive populations into decline (Bergerud 1967, 1974).

Of the models evaluated, total disturbance, expressed as proportional amount of range affected, was the best predictor of observed recruitment levels in caribou, explaining 61% of the variation in this parameter. An assumption implicit in the use of a simple model is that areas within population ranges or study areas that are not burned or impacted by anthropogenic features are equally good for caribou, which may or may not be the case. Exploring the variability in response across ranges, closer examination of the specifi c conditions on individual ranges, and consultation with biologists familiar with local circumstances, could help to identify reasons underlying populations falling outside the confi dence intervals of the regression, and generate additional hypotheses about measures affecting range condition for evaluation in future analyses. An obvious additional attribute of disturbance that could be quantifi ed using existing data is the spatial confi guration of disturbances within caribou ranges, and their effect on measures of connectivity and patch size. There exists both theoretical and empirical evidence to suggest that, at the same level of disturbance, a more dispersed spatial pattern would lead to greater fragmentation of the range, greater interspersion of high quality caribou habitat with that suitable for other species, increased accessibility of the range by predators, and thus an overall decrease in available refuge areas for caribou, leading to negative effects on population condition.

The measure of population condition employed in this study was recruitment, for which the most extensive data set was available. Exploratory analyses revealed good correspondence between recruitment and population growth for a subset of the available data. However, recruitment was not correlated with female survival, as suggested for caribou populations in previous studies (e.g., Bergerud 1988). We had earlier hypothesized that a disjunct might exist. Future analyses should explore the relationship between recruitment and other population parameters through empirical and simulation studies. To be of greatest utility to management, demographic analyses should focus on the co-variation between vital rates and habitat variables (Boyce et al. 2005), in this case measures of range condition. There are several important outcomes from such work. First, it would increase understanding of the relationship between the components of population growth and their interaction with range condition, and identify uncertainties that could become the focus of future adaptive management experiments. Second, it would inform monitoring schemes for caribou, such that the data collected represent the most cost-effi cient and effective measures of population condition. The development of long-term, standardized monitoring programs and protocols would produce consistent estimates that maximize the information available for future analyses.

Previous work suggests that population response may lag behind landscape change by up to several decades, due to the proximate factors responsible (Vors et al. 2007). Effects on caribou populations mediated by changes in competitors and predators can take some time to emerge, as numerical response by these species is not be immediate. Our analyses did not address potential time lags in population response to changing range condition, as the Global Forest Watch Canada (GFWC) anthropogenic disturbance data could not be partitioned into time intervals. However, GFWC is presently completing a landscape change analysis, which quantifi es anthropogenic changes over the time intervals 1990-2000, and 2001-2007. These data will facilitate investigation of caribou population dynamics relative to rates of change, as well as exploration of potential time lags in response.

A primary objective of the present study was to extend the Sorensen et al. (2008) analysis to a broader range of population and landscape conditions. The general model structure employed for each study was similar; however, different measures of both the independent and dependent variables were evaluated. Thus, it is not appropriate to quantitatively compare specifi c model outputs. Nevertheless, both studies posed the question: is there a relationship between human-caused disturbance and caribou population performance? The answer is affi rmative. There is an increasing risk to caribou population persistence as the level of anthropogenic disturbance increases, and disturbance by fi re interacts with this, such that the total disturbance on a caribou range must be considered when developing management guidelines. The results further suggest that it is possible to establish quantitative guidelines for disturbance thresholds relative to probability of population persistence, even though the mechanisms underlying the relationship may not be fully understood. Ultimately, the evaluation and management of habitat must be tied to demographic responses, like recruitment. Assembling and analyzing information from multiple populations – the product of many years of effort from many individuals - is one means to generate such vital knowledge.

 

Page details

Date modified: