Automated Classification of Morphologically Identical Mosquito Sibling Species Using Wingbeat Harmonics

0.00 0.01 0.02 0.03 0.04 0.05 0.06

-200

00

1000

A. Waveform

time (s)

ampl

itude

(bi

ts)

0 1000 2000 3000 4000

0.0

0.4

0.8

B. Frequency spectrum

frequency (Hz)

ampl

itude

(ar

bitr

ary

units

)

w ingbeatfrequency363.3 Hz

C. Signature

harmonic number

ampl

itude

(ar

bitr

ary

units

)

1 2 3 4 5 6 7 8 9 10

Aubrey Moore1, Charles Apperson2, John McLaughlin3, Philipp Kirsch3, Dariusz Czokajlo3

1College of Agriculture & Life Sciences, University of Guam, Mangilao, Guam; 2Department of Entomology, North Carolina State, Raleigh University, NC; 3APTIV, Inc., Portland, OR

INTRODUCTIONWingbeat frequency has been used as a character for differentiating insect species. Here, we show that harmonic patterns associated with wingbeat frequency provide additional species-specific information which can be used to differentiate closely related species with overlapping wingbeat frequencies.Here, we report how we used harmonic patterns to differentiate female Culex pipiens mosquitoes from Cx. quinquefasciatus. These sibling species are morphologically identical. Previously, they could be differentiated only by molecular methods.

Figure 1. Origin of harmonics in optically-sensed wingbeat waveforms. Images are from a web page entitled “How Insects Fly” (http://park.org/Canada/Museum/insects/flight/flapping.html). To simulate data from a non-imaging photosensor, we calculated the mean sum of the red, green, and blue values for pixels in each image. This values were plotted as wingbeat waveforms. Fast Fourier transform of these waveforms generate very different spectra, even though wingbeat frequencies are identical. Figure 3. Digital signal processing.

Figure 2. Apparatus.

METHODSWingbeat waveforms were recorded using FAST-ID instrumentation developed by APTIV Inc. Waveforms were recorded when female mosquitoes from two colonies of Cx. pipiens (CT, MA) and two colonies of Cx. quinquefasciatus (CA, FL) flew between an infrared LED array and a miniature solar cell (Fig. 2). For each waveform, wingbeat frequency was estimated using a YIN estimator, frequency spectrum was calculated using a fast Fourier transform, and a harmonic pattern was derived by integrating area under the frequency spectrum adjacent to each harmonic (Fig. 3).

WingbeatFrequency

(Hz)

Cx. pipiens Cx. quinquefascatusCT MA CA FL

321–333 10 1 4 1334–348 16 10 12 4349–364 27 73 21 1365–381 67 151 7 2382–400 67 173 28 11401–421 40 18 45 25

Table 1. Number of waveforms in each frequency bin.

RESULTSWingbeat frequencies of female Cx. pipiens and Cx. quinquefasciatus were not different (Table 1). However, harmonic patterns for waveforms within each frequency bin were significantly different (Fig. 4). Cx. quinquefasciatus produced stronger higher order harmonics than Cx. pipiens.In a cross-classification test with a nearest neighbor classifier using wingbeat frequencies and harmonic patterns, 65% of the mosquitoes were classified correctly, 31% were unclassified, and only 4% were misclassified.

Figure 4. Harmonic patterns. Each bar represents relative energy associated with each harmonic. Error bars are SEM. Filled circles indicate significant differences.

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 11

Cx. quinquefasciatus n = 5

Wingbeat frequency = 321 to 333 Hz

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 218

Cx. quinquefasciatus n = 9

Wingbeat frequency = 365 to 381 Hz

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 26

Cx. quinquefasciatus n = 16

Wingbeat frequency = 334 to 348 Hz

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 240

Cx. quinquefasciatus n = 39

Wingbeat frequency = 382 to 400 Hz

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 100

Cx. quinquefasciatus n = 22

Wingbeat frequency = 349 to 364 Hz

H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11

Cx. pipiens n = 58

Cx. quinquefasciatus n = 70

Wingbeat frequency = 401 to 421 Hz

Wingbeat Waveform

Wingbeat Spectrum

Harmonic Pattern