Appendix 1. Range, mean and standard deviation (SD) of, and correlation (Pearson’s r) between, forest amount and number of forest patches across 22 landscapes at multiple spatial scales
Scale / Forest Amount (ha) / Number of Patches / rRange / Mean / SD / Range / Mean / SD
1 km / 13.32–134.22
(4.24–42.72%) / 52.25 / 32.40 / 4–19 / 10.14 / 4.51 / .392
1.5 km / 39.73–334.17
(5.62–47.28%) / 136.73 / 81.62 / 8–41 / 21.45 / 8.39 / .175
2 km / 63.02–574.69
(5.01–45.73%) / 266.55 / 136.79 / 10–61 / 34.27 / 13.09 / .342
2.5 km / 82.94–832.20
(4.22–42.38%) / 438.05 / 210.75 / 15–84 / 47.14 / 17.35 / .186
3 km / 106.08–1233.25
(3.75–43.62%) / 700.86 / 338.61 / 25–139 / 71.68 / 28.39 / .132
4 km / 572.91–3964.01
(11.40–78.86%) / 2497.96 / 1075.15 / 49–268 / 111.09 / 48.04 / .080
5 km / 785.00–3827.71
(9.99–48.74%) / 2668.66 / 992.38 / 94–365 / 171.27 / 65.05 / .061
Appendix 2. Range, mean and standard deviation (SD) of local predictor variables measured in the 22 focal patches
Local Variable / Range / Mean / SDSampling date (Julian) / 1.00–63.00 / 30.50 / 20.05
Tree density (#/ha) / 300–933.33 / 620.45 / 150.13
Snag density (#/ha) / 0.00–316.67 / 125.76 / 89.05
Appendix 3. Layout of bat recorders (■), insect light traps (●) and 10 x 10 m vegetation survey quadrats (- -) at each focal patch. Arrows (◄) indicate the direction in which the bat recorders were facing. In each 10 x 10 m quadrat, we counted the number of trees and snags > 2 m tall and > 10 cm dbh to quantify tree and snag density. Insect light traps consisted of a 15-Watt black light fitted with three vertical plastic veins, which was placed atop a reflective cone leading into a 19 L bucket. In each bucket, we used a strip of dichlorovos (Vapona) as a killing agent.
Appendix 4.Identification of bat recordings
The echolocation signals produced by bats consist of a sequence of discrete calls emitted in rapid succession. This entire sequence of calls is referred to as a “pass”. The structure of the calls within a pass will vary whether a bat is searching for, approaching, or attacking a potential target (Fenton and Bell 1981). When a bat commutes or searches for prey, it produces what are known as “search-phase” calls (Griffin et al 1960). Once a potential target is identified, the pulse repetition rate increases (termed “approach phase”) before ending in a terminal or “feeding buzz” (Griffin et al 1960). Since the duration and structure of calls vary based on the task, search-phase calls are usually the most consistent and identifiable calls recorded.
We used quadratic discriminant function analysis (DFA) to classify our recordings to species by comparing the parameters of our recorded calls to a library of validated reference calls (Russo and Jones 2002). To build the species identification model, we started with a library consisting of 269 full-spectrum recordings from free-flying bats at various locations in Ontario, Canada, including: P. subflavus (n = 33), M. septentrionalis (n = 21), M. lucifugus (n = 68), L. noctivagans (n = 28), E. fuscus (n = 51), L. borealis (n = 29), and L. cinereus (n = 39) (Hooton and Adams, unpublished). We scanned all reference recordings and retained only those passes containing two or more calls (Thomas 1988). We then selected for analysis the one search-phase call from each pass with the highest signal-to-noise ratio (Parsons and Jones 2000). We then measured nine frequency and time parameters from each of these reference calls: call duration; inter-pulse interval (time between successive calls in a pass); maximum frequency; minimum frequency; bandwidth (difference between maximum and minimum frequency); frequency with most energy; characteristic frequency (part of call with maximum slope change); global slope (linear function characterizing call shape); and curvature (polynomial function characterizing call shape). We used SCAN’R 4.41 (Binary Acoustic Technology) to process acoustic recordings and measure call parameters.
While the absence of collinearity among variables is not an assumption of DFA, highly correlated variables may bias the results (Quinn and Keough 2002). We therefore used variance inflation factors (VIF) to detect collinearity among call parameters, removing those with a VIF > 10 (Neter et al 1990). As a result, we removed ‘bandwidth’ and ‘characteristic frequency’ from the analysis. We then used a MANOVA test to ensure that the remaining seven call parameters provided significant species discrimination. We tested for multivariate normality by examining Q-Q plots and Shapiro-Wilk tests for each variable and found that the data were slightly non-normal. However, multivariate analyses are relatively robust to deviations from normality (Quinn and Keough 2002). Since the data failed to meet the homogeneity of variance test for multivariate analyses (Box’s M = 1887.23, F = 10.22, d.f. = 168, p < .001), we used quadratic DFA (Parsons and Jones 2000; Russo and Jones 2002; Preatoni et al 2005). We then quantified model classification rates using leave-one-out cross-validation, which provides an unbiased reclassification error estimate (Preatoni et al 2005).
The quadratic DFA model reached an overall correct classification rate of 88.8% (239 of 269 reference calls correctly classified). Correct classification rates for each species were: P. subflavus 87.9%; M. septentrionalis 95.2%; M. lucifugus 79.4%; L. noctivagans 92.9%; E. fuscus 90.2%; L. borealis 93.1%; and L. cinereus 94.9%. A MANOVA also showed that the model provided significant species discrimination (Wilk’s = .664, F = 18.844, d.f. = 7, p < .001).
To identify our own recordings to species, we then queried the DFA model using the same seven parameters that were used to build the model. We used posterior probabilities as a measure of confidence for each individual classification. If the posterior probability for a classification was < .95, or if < 5 passes from a single species were identified at a given site, the classification was verified by entering a second random call selected from the original pass into the DFA model. If there was uncertainty or inconsistency in the classification of the second call, that particular pass was considered unidentifiable and labelled as ‘unknown’. We used Minitab 15 (2006) for the quadratic DFA and species classifications.
References
Fenton MB, Bell GP (1981) Recognition of species of insectivorous bats by their echolocation calls. J Mammal 62(2):233–243
Griffin DR, Webster FA, Michael CR (1960) The echolocation of flying insects by bats. Anim Behav 8:141–154
Neter J, Wasserman W, Kutner MH (1990) Applied linear statistical models, 3rd edn. Irwin, Chicago
Parsons S, Jones G (2000) Acoustic identification of twelve species of echolocating bat by discriminant function analysis and artificial neural networks. J Exp Biol 203:2641–2656
Preatoni DG, Nodari M, Chirichella R, Tosi G, Wauters LA, Martinoli A (2005) Identifying bats from time-expanded recordings of search calls: comparing classification methods. J Wildl Manage 69(4):1601–1614
Quinn G, Keough M (2002) Experimental design and data analysis for biologists. Cambridge University Press, Cambridge
Russo D, Jones G (2002) Identification of twenty-two bat species (Mammalia: Chiroptera) from Italy by analysis of time-expanded recordings of echolocation calls. J Zool 258:91–103
Thomas DW (1988) The distribution of bats in different ages of Douglas-fir forests. J Wildl Manage 52(4):619–626