2005 Young Naturalist Awards - The World Through a Bat's Ears

Sarah The World Through a Bat's Ears

Continued...

< 1 | 2 | 3 | 4 | >

Materials and Methods

Sarah with recording equipment

I recorded the bats' echolocation calls with an Avisoft-Ultrasoundgate 116 microphone operated with a Dell Latitude D800 PC and Avisoft-USG software on 17 August 2004 at the Renfrew Mine in Ontario, Canada.

In each setting, the microphone was mounted on a tripod, pointed at 60 degrees up from the horizontal. I recorded the bats' calls in four settings with different degrees of clutter: (a) over an open field, where the nearest clutter (trees) was 6.3 meters away;

Entrance to abandoned mine

(b) in vegetation, where the nearest branches and leaves were 0.7 meters away; (c) at the entrance to an abandoned mine to record approaching bats; and (d) within the mine adit (the nearly horizontal passage from the surface into the mine). In (c), I recorded the calls of bats as they approached the entrance to the mine, which was 1.7 m wide by 2.6 m high, while in (d) I recorded bats within the mine adit, which was 1.8 m wide by 2.8 m high, and 1.5 meters from a gate composed of a series of 2 cm bars every 16-17 cm, an obstacle they had to navigate in order to move deeper into the mine.

Chart 1: Bat call waveforms (Click to enlarge)

In the laboratory, I reviewed my hundreds of recordings and selected suitable calls for analysis. Calls selected were those where the signal was at least 10% stronger than the background.

I then analyzed the calls using BatSoundPro software. For each of the four settings, I analyzed three sequentially produced echolocation calls taken from three call sequences. Call sequences were at least 60 seconds apart. I measured call duration (Dur) and inter-pulse interval (IPI) from the time amplitude and spectrogram displays (Chart 2).

Chart 2: Bat calls recorded in different settings (Click to enlarge)

Power spectra were used to produce three measures of sound frequency (in kHz): lowest frequency (LF), frequency of maximum energy (FME), and highest frequency (HF). Frequency of maximum energy (most intense frequency or peak frequency) is the frequency that exhibits the greatest relative power on the power spectrum. I measured the frequency of maximum energy from the fast Fourier transform (FFT) power spectrum, the highest and the lowest frequency from the FFT at -10 dB from FME (Chart 3). These data were recorded in a Microsoft Excel data sheet which I designed.

Chart 3: Bat call in fast Fourier transform (FFT) display (Click to enlarge)

I then statistically analyzed my data using statistical software SPSS 12.0, following Biscardi et al (in press). I transferred the call data from Microsoft Excel to SPSS. I then used a Multiple Analysis of Variance (MANOVA) to assess the level of variation in the call features I had measured (Dur, FME, HF, LF and IPI) according to their settings. Since there was significant variation, I proceeded to use Discriminate Function Analyses (DFA) to assess the accuracy with which bat calls could be classified into the different clutter settings. I then proceeded with a DFA to determine the parameters that were most important in identifying the calls, as well as the accuracy with which the calls could be classified as coming from a particular setting. Lastly, I examined the cross-validated classification results. (See Appendix 1: Note About Statistics.)

Results

I analyzed a total of 36 echolocation calls of M. lucifugus, nine for each of the four settings. The descriptive statistics are presented in Table 1. Calls produced in the open setting are longer in duration, and are produced at longer inter-pulse intervals and at lower frequencies, than calls produced in the three closed (cluttered) settings. Among the cluttered settings, calls produced in the forest vegetation are longer in duration, and are produced at longer inter-pulse intervals and at lower frequencies, than calls produced in or near the more restricted environment of the mine. The results from the multiple analysis of variance (MANOVA) (Table 2) revealed that the variation in call features according to setting was significant. This level of variation was validated using discriminate function analyses (DFA) to classify the data according to setting.

Table 1. The descriptive statistics (mean + standard deviation) of the echolocation calls I recorded and analyzed. Presented here are sample size (N), duration (Dur), frequency of maximum energy (FME), highest frequency (HF), lowest frequency (LF), and inter-pulse interval (IPI). The recordings were made in an open field, in vegetation, at the entrance of the mine, and within the mine adit.

Setting	N	Dur (ms)	FME (kHz)	HF (kHz)	LF (kHz)	IPI (ms)
Open Field	9	8.3 + 2.1	35.4 + 8.9	44.0 + 9.2	32.8 + 6.8	152.0 + 78.4
Vegetation	9	3.0 + 0.5	47.6 + 3.4	51.5 + 3.8	44.0 + 3.3	70.9 + 13.0
Entrance of Mine	9	2.1 + 0.6	53.3 + 6.2	62.3 + 7.2	45.3 + 4.3	38.0 + 19.7
Inside Mine	9	2.3 + 0.3	52.3 + 2.0	64.5 + 12.2	47.3 + 4.0	61.2 + 19.7

Table 2. The results of MANOVA analyses of the data from the echolocation calls reveal significant variation in call features according to the setting (degree of clutter) in which the calls were recorded. Here, "all" means all the call features measured, Dur, FME, HF, LF, and IPI. For each F value, the first subscript number is the hypothesis degrees of freedom, and the second is the error degree of freedom.

	Call Features	Wilks' Lambda	F-value	Probability
Setting	All	0.003	F5, 16 = 980	< 0.001
Setting	Dur, FME	0.004	F2, 31 = 4074	< 0.001

Having determined there was significant variation, I used a series of DFA analyses to explore changes in bat echolocation calls according to clutter (Tables 3a-3d). The first analysis (Table 3a) using all call features showed 75% correct classification of bat calls by clutter setting. The DFA analysis revealed that Function 1 (of 3) accounted for 97.7% of the variation, and Function 1 largely reflected the contributions of two call features, duration and frequency of maximum energy. Dur and FME, therefore, were the two most important parameters distinguishing the four settings. I then used only duration and FME for another DFA (Table 3b), and the level of correct classification fell to 55.6%. This indicates that other call features besides Dur and FME add to the precision of the DFA classification, and supports the view that bats are controlling all the features or the whole signal. It is interesting to note that although the DFA analysis sometimes failed to classify calls correctly among the different closed settings, a closed call was never classified as an open call, and an open call was only rarely classified as a closed call. Where an open call was classified as a closed call, it was always classified as a call in the vegetation setting and never as a call at the entrance to or within the more restricted environment of the mine. Amongst the closed calls, the level of correct classification was higher—83.3% in Table 3a and 77.8% in Table 3b—in the vegetation setting than at the entrance or within the mine.

< 1 | 2 | 3 | 4 | >