ECOLOGY AND BEHAVIOR Acoustic Detection of rhinoceros (Coleoptera: : ) and Nasutitermes luzonicus (Isoptera: Termitidae) in Palm Trees in Urban Guam

1 2 R. W. MANKIN AND A. MOORE

J. Econ. Entomol. 103(4): 1135Ð1143 (2010); DOI: 10.1603/EC09214 ABSTRACT Adult and larval Oryctes rhinoceros (L.) (Coleoptera: Scarabaeidae: Dynastinae) were acoustically detected in live and dead palm trees and logs in recently invaded areas of Guam, along with Nasutitermes luzonicus Oshima (Isoptera: Termitidae), and other small, sound-pro- ducing invertebrates and invertebrates. The low-frequency, long-duration sound-impulse trains produced by large, active O. rhinoceros and the higher frequency, shorter impulse trains produced by feeding N. luzonicus had distinctive spectral and temporal patterns that facilitated their identiÞcation and discrimination from background noise, as well as from roaches, earwigs, and other small sound-producing organisms present in the trees and logs The distinctiveness of the O. rhinoceros sounds enables current usage of acoustic detection as a tactic in GuamÕs ongoing O. rhinoceros eradication program.

KEY WORDS invasive , acoustic detection, eradication

Adult Oryctes rhinoceros (L.) Coleoptera: Scar- Spot treatment of palms and breeding sites has been abaeidae: Dynastinae) bore into and feed inside conducted with conventional and injectable insecti- emerging fronds at the crowns of (Cocos cides (Smith and Moore 2008). nucifera L.) and other palm trees in Southeast Asia Visual inspection of frond and trunk damage is the and tropical PaciÞc islands, causing extensive dam- most commonly used method of detecting O. rhi- age (Gressitt 1953, Hinckley 1973, Bedford 1980, noceros adults; however, infestations cannot be de- Jackson and Klein 2006). Frequent attacks by sev- tected easily from the ground until well after sig- eral adults may kill a tree. Larvae feed inside rotting niÞcant damage has occurred. Because adults, logs of palms killed by adults or by high winds from larvae, and pupae are known to produce audible typhoons that pass through the tropical PaciÞc an- sounds (Gressitt 1953, Mini and Prabhu 1990), it was nually. Introductions of O. rhinoceros into the Palau of interest to determine whether adults feeding and Islands and other originally uninfested regions have moving in the crowns of live trees and larvae feeding resulted in considerable economic harm to coconut in standing, dead trunks can be detected by acoustic palm plantations (Gressitt 1953). methods (e.g., Mankin et al. 2000, 2002, 2008a,b). O. rhinoceros was unknown in Guam until 2007 Previous experience with detection of sounds (Smith and Moore 2008), and due to the considerable in trees (e.g., Mankin et al. 2008b) suggested that a potential for economic damage, the Tumon Bay hotel combination of spectral and temporal pattern fea- district was quarantined a few days after an adult female was discovered in the area. Mass trapping was tures of detected sounds could be used to distin- initiated with commercially available aggregation guish O. rhinoceros signals from nontarget signals in pheromone oryctalure (ethyl 4-methyloctanoate) the noisy urban areas of the quarantine zone. A large (Hallett et al. 1995, Ramle et al. 2005), and a program insect like O. rhinoceros might perform boring and was established to remove or treat rotting coconut logs feeding activities that produced louder and longer and other potential breeding and larval feeding sites. sounds than those produced by other small that also were associated with palm trees, including Pycnoscelus surinamensis (L.) (Blattodea: Blaberi- The use of trade, Þrm, or corporation names in this publication does dae), Brontispa palauensis (Esaki & Chujo) (Co- not constitute an ofÞcial endorsement or approval by the United States Department of Agriculture, Agricultural Research Service of leoptera: Chrysomelidae) (Muniappan 2002), Che- any product or service to the exclusion of others that may be suitable. lisoches morio (F.) (Dermaptera: Chelisochidae), 1 Corresponding author: USDAÐARS, Center for Medical, Agricul- and tenebrionids. Here, we report on a study con- tural, and Veterinary , Gainesville, FL 32608 (e-mail: ducted to determine the feasibility of incorporating [email protected]). 2 College of Natural and Applied Sciences, University of Guam, acoustic detection methods into the ongoing pro- Mangilao, Guam 96923. gram to isolate and eradicate breeding populations. 1136 JOURNAL OF ECONOMIC ENTOMOLOGY Vol. 103, no. 4

Materials and Methods signal below 200 Hz was background noise and no important signal features were observed Ͼ5 kHz, the Records were collected over an 11-d period from 11 signals were band-pass Þltered between 0.2 and 5 kHz live palm trees, six dead palm trees, seven logs, and Þve to facilitate subsequent analyses. In the initial screen- stumps that were easily accessible and available for ings, we conÞrmed the presence of groups (trains) of (destructive) inspection at several different locations discrete, 3Ð10-ms impulses separated by intervals in the Tumon Bay hotel district quarantine zone. Ex- Ͻ250 ms. Such trains had appeared frequently in re- ternal veriÞcation of the insects present at each site Ϸ cordings at sites where insects were recovered in pre- was performed within 24 h after recording by felling vious studies (Zhang et al. 2003a,b; Mankin et al. a tree if it was standing, splitting open the trunk, log, 2008a,b). Trains containing 15 or more impulses were or stump tissue with axes and knives, and sorting a focus of analysis because they often were identiÞed through the separated pieces. The surveyed locations as insect sounds in playbacks of recordings from in- included the grounds of a hospital, a hotel, a retire- fested sites, but trains containing only small numbers ment complex, a shopping center, and three houses, as of impulses, as well as signals without any 3Ð10-ms well as a rubbish pile. Although O. rhinoceros were impulses, usually were interpreted as background known to be present in the quarantine zone, none of noise (Mankin et al. 2008b). the individual trees, logs, or stumps in the Þeld study The impulses and impulse trains detected in the had been inspected previously. Temperatures were recordings were analyzed with customized software, Ϸ Њ 28Ð31 C at the times of recording. DAVIS (Digitize, Analyze, and Visualize Insect To gain additional knowledge of the characteristics Sounds, Mankin 1994, Mankin et al. 2000), which dis- of signals produced by larval and adult O. rhinoceros, carded long-duration, low frequency background records were collected also from separate groups of O. noise (Mankin et al. 2007, 2008a,b) and then compa- rhinoceros larvae and adults maintained in a rearing red the spectrum of a 512-point time-slice centered facility at the University of Guam. These were around the peak of each impulse against averaged held in 30- by 24- by 24-cm plastic rearing cages con- spectra (spectral proÞles) of independently veriÞed taining pieces of sugar cane and palm trunk. The lab- insect-produced signals and impulsive, tree bending- oratory studies are described separately from the Þeld grinding noises (see next section). The spectral com- studies in the last section of the Results because the parisons were performed after normalizing the accel- spectral characteristics of the sounds depend strongly eration (Mankin and Benshemesh 2006). on the substrate in which insects are moving and Insect and Background-Noise Spectral Profiles. feeding (Mankin et al. 2008). Spectral proÞles (Mankin 1994, Mankin et al. 2000) Acoustic Sensors and Recording Procedures. Signals were constructed from background noise impulse were collected from accelerometers attached to trains and four types of insect sound impulse trains, charge ampliÞers (Mankin et al. 2000, 2001, 2002). The two of which were produced by the primary targets, ampliÞed signals were saved on a dual-channel, digital adult and larval O. rhinoceros. An adult O. rhinoceros audio recorder sampling at 44.1 kHz (24 bits). In proÞle, rhino_a, was averaged from 33 consecutive recordings at a survey location, the accelerometers impulse trains collected at a live palm where an adult were attached magnetically to 30-cm-long spikes, in- was recovered, and a larval proÞle, rhino_l, from 35 serted into the wood a few minutes before recording. consecutive impulse trains collected at the log where In laboratory recordings of O. rhinoceros held in rear- 72 larvae were recovered. One of two nontarget-insect ing containers, the accelerometer was attached to the proÞles, termite, was constructed from 233 consecu- rearing container or to a small nail inserted into one tive impulse trains produced in a dead palm where of the palm pieces on which the insects were feeding. only Nasutitermes luzonicus Oshima (Snyder and Listeners (see Acknowledgments) could monitor the Francia 1960, Su and Scheffrahn 1998) were recov- signals using headphones attached to the recorder, ered, and the second from 45 consecutive impulse thereby avoiding levels of wind or street noise high trains produced in a live palm where two nontarget enough to mask insect sounds. Wind noise caused species, P. surinamensis and B. palauensis, were the delays in 20Ð30% of the recording trials, but in all cases only organisms recovered. A background noise proÞle there were sufÞcient intervals of reduced background was generated as a spectral average of wind-induced noise to permit useful recordings of insect sounds for tapping and trunk bending and grinding that produce periods of 45 s or longer. In the absence of background short, broadband sound impulses (e.g., Fukuda et al. interference, records were collected for 180 s or 2007). A series of 38 consecutive impulse trains re- longer. Dual-channel recordings from two accelerom- corded at an uninfested palm were used in construct- eters were obtained from several Þeld sites, in which ing this proÞle. cases the channel with the greatest clarity was se- After the proÞles mentioned above were con- lected for analysis. structed from individual Þles, the complete set of Þles Digital Signal Processing and Classification. After was reviewed in DAVIS, which discarded impulses the recording sessions each day were completed, the from further analysis if their spectra failed to match recorded signals were transferred to a laptop com- the rhino_a, rhino_l, or termite spectral proÞles within puter and initial analyses of their spectral and tem- an empirically determined difference threshold, Ts, or poral characteristics were conducted using Raven 1.3 if they matched one of the two nontarget proÞles more software (Charif et al. 2008). Because much of the closely than any insect sound proÞle. The occurrence August 2010 MANKIN AND MOORE:INVASIVE O. rhinoceros ACOUSTIC DETECTION 1137 times of impulses that matched an insect sound proÞle Table 1. Names of indicator (A) and rate (B) variables for O. (valid impulses) were saved in an impulse-sequence rhinoceros and N. luzonicus trains and bursts, and values of cutoff spreadsheet and labeled according to which of the rates (C), below which infestations rarely occurred (lower) and above which infestations were likely (upper), estimated from the observed three spectral proÞles they matched best. distributions of O. rhinoceros and termite trains and bursts at Temporal Pattern Analyses. The impulse sequences infested and uninfested sites (see Results) were screened to identify and characterize groups (C) Cutoff (trains) of impulses that listeners typically classify as (A) Indicator (B) Rate variable (unit) rates (no./min) separate, individual sounds, i.e., groups of impulses variable separated by intervals Ͻ250 ms (Mankin et al. 2008b). lower upper Each train was labeled according to the spectral pro- irhino_l-t rrhino_l-t (trains/min) 25.0 50.0 Þle matched by a plurality of its impulses. The begin- irhino_l-b rrhino_l-b (bursts/min) 0.5 5.0 i r (trains/min) 15.0 35.0 ning and ending times of impulse trains, their labels, rhino_a-t rhino_a-t irhino_a-b rrhino_a-b (bursts/min) 0.5 2.0 and the number of impulses per train, nt, were stored itermite-t rtermite-t (trains/min) 15.0 35.0 in separate train-sequence spreadsheets for each re- itermite-b rtermite-b (bursts/min) 5.0 10.0 cording. Ն In each row, the indicator variable value is high, for rate upper Train-sequence spreadsheets were screened fur- Ͼ Ն cutoff; medium, for upper cutoff rate lower cutoff; and low, ther to determine the range of impulse counts found otherwise, where the labels “rhino” refer to O. rhinoceros rates and in trains of each proÞle. Initially, we examined the cutoffs, “termite” to termite rates and cutoffs, “_l” to larvae, “_a” to adults, “t” to trains, and “b” to bursts. distributions of nt in recordings where only O. rhinoc- eros was present at a site, and trains with Ͼ10 impulses were played back through the audio feature in Raven to determine whether a listener assessed them as in- more, we considered two comprehensive indicators of sect sounds, background noise, or unclassiÞable. Based O. rhinoceros infestation, combining the cases of larval and adult infestation, i.e., on these screenings, minimum counts, nmin-rhino_a and ϭ nmin-rhino_l, were estimated for trains with sufÞcient irhino-t impulse counts to be classiÞed as insect sounds, des- h if i _ ϭ h or i _ ϭ h ignated hereafter as bursts (Mankin et al. 2008a,b). igh rhino l-t igh rhino a-t igh Because termites were found at many recording medium if irhino_l-t ϭ medium and irhino_a-t Ͻ high sites, we conducted a similar analysis for sites where ϭ Ͻ only termites were present, and a minimum count, Ά or irhino_a-t medium and irhino_l-t high · nmin-termite, was estimated to classify termite trains of sufÞcient size as termite bursts. low otherwise, with indicator values speciÞed as in We then reprocessed the complete set of train- [1] sequence spreadsheets. Trains labeled as rhino_a or _ Ͼ Ͼ Table 1, and similarly, rhino l, with nt nmin-rhino_a or nt nmin-rhino_l, were ϭ set as rhino_a or rhino_l bursts, respectively. Trains irhino-b labeled as termite, with n Ͼ n , were set as t min-termite h if irhino_l b ϭ h or irhino_a b ϭ h termite bursts. These procedures were similar to those igh - igh - igh used previously to specify bursts produced by hidden medium if irhino_l-b ϭ medium and irhino_a-b Ͻ high larvae of two other insect species, Anoplophora [2] ϭ Ͻ glabripennis (Motschulsky) (Mankin et al. 2008b) and Ά or irhino_a-b medium and irhino_l-b high · Rhynchophorus ferrugineus (Olivier) (Mankin et al. 2008a). low otherwise.

Computer-Rated Likelihood of Infestation. Based Finally, the distributions of high,medium, and low on previous occurrences of strong associations be- likelihood of infestation of O. rhinoceros or termites at tween insect infestations and trains or bursts of insect recording sites were compared at infested and unin- sound impulses (Mankin et al. 2008a,b), we consid- fested sites using the Wilcoxon two-sample exact test ered the possibility that such associations occurred (Proc NPAR1WAY, SAS Institute 2004) under the null also for O. rhinoceros infestations and rhino_l or rhi- hypothesis that the distributions of the likelihood of no_a trains or bursts, as well as for N. luzonicus infes- infestation were independent of whether the sites tations and termite trains or bursts. We therefore con- were infested or uninfested. structed indicators of infestation likelihood (Table 1) by using procedures similar to those described in Man- Results kin et al. (2007). First, cutoffs were estimated for rates of trains and bursts, lower, below which infestations Three of 29 recording sites in the quarantine zone rarely occurred, and for rates, upper, above which in- were found to contain O. rhinoceros. Eight other sites festations were likely (see column C, Table 1). The were free of sound-producing organisms. No O. rhi- cutoffs were estimated from the observed distribu- noceros were found at the remaining sites, but other tions of O. rhinoceros and termite train and burst rates sound-producing organisms were present, including at infested and uninfested sites. The indicator values B. palauensis, C. morio, P. surinamensis, tenebrionids, for O. rhinoceros and termite trains and bursts were set diplopodans, tree frogs, and geckos, hereafter denoted as high,medium,orlow, as denoted in Table 1. Further- as nontarget organisms. Nasutitermes luzonicus, found 1138 JOURNAL OF ECONOMIC ENTOMOLOGY Vol. 103, no. 4

Fig. 1. Oscillogram (A) and spectrogram (B) of signals recorded by accelerometer from an O. rhinoceros adult in a palm tree, with an individual sound impulse expanded in the inset. Signals enclosed by dotted lines indicate a long series (train) of impulses that was classiÞed by computer analysis as an O. rhinoceros burst. The 42 impulses included in the burst are noted with dots. Three subsequent short impulse trains are underlined below the time scale. Darker shades in the spectrogram indicate frequencies with higher signal energy at the speciÞed time. (Online Þgure in color.) in Þve dead trees and logs that did not contain O. sects, P. surinamensis and B. palauensis, and a proÞle rhinoceros, were among the most abundant nontarget for background noise signals produced by wind- and sound-producing organisms. tree-grinding. A 10-s sample of signals recorded at a site with an Temporal Pattern Analysis. To consider the possi- adult O. rhinoceros is shown in Fig. 1A and B. The bility that O. rhinoceros might perform activities pro- sample contains multiple, 3Ð10-ms impulses, one of ducing louder and longer sounds than from roaches which is expanded in the inset of Fig. 1A. Four impulse and other small insects in the palm trees, we examined trains occur in the sample. The Þrst train, classiÞed by the distribution of numbers of valid impulses in trains the DAVIS signal analysis program as a rhino_a burst, recorded from sites where only O. rhinoceros was includes 42 impulses that are marked by dots in the recovered (three sites), and also examined the distri- dashed rectangle. Three subsequent, shorter trains are bution of impulses at sites containing only nontarget underlined below the time scale. The impulses were organisms (14 sites). The fraction of O. rhinoceros broad-band signals with dominant peaks below Ϸ2 trains with high impulse counts was greater for re- kHz and secondary peaks between 4 and 5 kHz (Fig. cordings in trees where O. rhinoceros was recovered 1B). Background noise is visible in Fig. 1B as a dark than for recordings in trees with only nontarget or- band extending through the sample interval, primarily ganisms. Considering trains where the majority of between 0 and 1.5 kHz, sometimes expanding to 2.5 impulses matched the rhino_l proÞle, for example, kHz during periods of high wind. 5.5% of 396 trains contained Ͼ20 impulses in O. rhi- Signals produced by O. rhinoceros larvae usually noceros recordings, compared with 2.5% of 1,440 trains were lower in amplitude, but similar in duration to in nontarget-organism recordings. In trains where the those produced by adults, as in the 10-s sample in majority of impulses matched the rhino_a proÞle, Fig. 2A and B, recorded from a palm log where 72 10.3% of 143 trains contained Ͼ20 impulses in O. rhi- larvae were recovered. The sample includes a short noceros recordings, compared with 0.5% of 1407 trains ϭ train (nt 13 impulses) underlined below the time in nontarget-organism recordings. We therefore set Ͼ ϭ scale, followed by a longer train that the DAVIS larval- or adult-proÞle trains with nt nmin-rhino_l _ ϭ ϭ _ _ program classiÞed as a rhino l burst (nt 28 im- nmin-rhino_a 20 impulses as rhino l or rhino a bursts, pulses marked by dots in the dashed box), an im- respectively (see Materials and Methods). pulse of which is expanded in the inset of Fig. 2A. Because N. luzonicus and other termites can be The larval and adult signals had similar spectral important pests in human structures (Acda 2007) and characteristics, shown in the spectral proÞles of Fig. were encountered frequently during the study, we 3 along with a spectral proÞle for N. luzonicus, a also examined the distributions of the numbers of proÞle for signals produced by two nontarget in- impulses in termite trains. In this case, the distribu- August 2010 MANKIN AND MOORE:INVASIVE O. rhinoceros ACOUSTIC DETECTION 1139

Fig. 2. Oscillogram (A) and spectrogram (B) of signals recorded by accelerometer from O. rhinoceros larvae in a palm log, with an individual sound impulse expanded in the inset. Signals enclosed by dotted lines indicate a group (train) of impulses that was classiÞed by computer analysis as an O. rhinoceros burst. The 28 impulses included in the burst are noted with dots. A shorter impulse train is underlined below the time scale. Darker shades in the spectrogram indicate frequencies with higher signal energy at the speciÞed time. (Online Þgure in color.) tions of termite trains were similar for infested and minimum count for a rhino burst, and thus shorter uninfested sites but the total numbers of termite in duration. trains with Ͼ15 impulses per train was much greater Distribution of O. rhinoceros Trains and Bursts at sites where N. luzonicus were present. The min- Among Infested and Uninfested Sites. The complete imum count for a termite burst was estimated at set of Þeld recordings was analyzed in DAVIS by ϭ nmin-termite 15 impulses, slightly fewer than the using the proÞles in Fig. 3 to consider the association

Fig. 3. Spectral proÞles of series of impulses recorded from an O. rhinoceros adult in a live palm tree (rhino_a, dashed line), O. rhinoceros larvae in a log (rhino_l, solid line), N. luzonicus in a dead tree (termite, dash-dot-dotted line), other organisms in a live palm (dash-dotted line), and wind and bending-grinding noises in a live tree (dotted line). Spectrum level is relative to the maximum acceleration measured in the 0.2- to 5-kHz reference range. (Online Þgure in color.) 1140 JOURNAL OF ECONOMIC ENTOMOLOGY Vol. 103, no. 4

Table 2. Rates (number per minute) of O. rhinoceros larval- Table 4. Rates (number per minute) of O. rhinoceros larval- profile trains, rrhino_l-t, and bursts, rrhino_l-b, adult-profile trains, profile trains, rrhino_l-t, and bursts, rrhino_l-b, O. rhinoceros adult- rrhino_a-t, and bursts, rrhino_a-b, and termite-profile trains, rtermite-t, profile trains, rrhino_a-t, and termite-profile trains, rtermite-t, and and bursts, rtermite-b, detected in palm trees or logs holding specified bursts, rtermite-b, detected in palm stumps from which nontarget numbers of O. rhinoceros larvae or adults, arranged in order of sound-producing organisms but no O. rhinoceros or termites were descending larval-profile burst rates, rrhino_l-b, at infested and un- recovered, ranked in order of descending rates of termite bursts, infested sites rtermite-b

Larval proÞle Adult proÞle Termite proÞle Termite No. Larval (rrhino_l) Adult (rrhino_a) No. (r _ ) (r _ ) (r ) (r ) rhino l rhino a termite (stage) termite insectsa Trains Bursts Trains Bursts Trains Bursts Trains Bursts Trains Trains Bursts 9.81 2.58 15.50 8.27 4.13 2.07 3 (larvae) 1.72 0.00 0.00 39.92 12.88 2 53.92 1.23 17.62 0.00 3.06 0.15 72 (larvae) 5.01 0.00 0.00 30.08 12.53 3 2.62 0.00 38.61 2.25 4.50 1.50 1 (adult) 6.90 1.70 0.50 25.19 9.50 3b 6.32 1.26 44.27 0.00 3.79 1.26 0.75 0.00 0.00 9.01 7.13 4 61.58 0.89 18.16 0.44 4.87 0.00 35.77 0.00 6.71 0.00 0.00 1 6.03 0.00 0.00 0.00 39.17 3.01 7.59 0.00 0.00 0.00 24.28 3.79 _ ϭ Note that the rate of rhino a bursts was rrhino_a-b 0.0 bursts/min 0.00 0.00 0.00 0.00 9.28 9.28 for all Þve stumps. 8.17 0.00 19.60 0.00 1.63 0.00 a B. palauensis, C. morio, and P. surinamensis. 24.55 0.00 0.00 0.00 13.75 1.96 b In addition, a tree frog and a gecko were recovered from this 6.28 0.00 3.14 0.00 28.25 9.42 stump. between the presence or absence of O. rhinoceros stumps, but rhino_l bursts were detected in one infestation and the rates of rhino_l and rhino_a stump (Table 4). trains and bursts. Although trains that matched ei- Given the positive association found between the ther the O. rhinoceros larval or adult proÞle ap- rates of O. rhinoceros bursts and the presence of O. peared at low rates in the recordings at uninfested rhinoceros at a recording site, we considered the con- sites (Table 2), bursts that matched these proÞles struction of indicators for automated rating of infes- were detected at only two of eight uninfested sites. tation likelihood, analogous to train- and burst-rate Trains and bursts that matched one or both proÞles indicators used successfully to detect infestations of were detected at rates Ͼ1 per min at all sites con- other insects in wood in previous studies (Mankin et taining O. rhinoceros. Rates of termite trains and al. 2008a,b). In constructing the indicators, lower and bursts were low at sites with O. rhinoceros and no upper cutoff rates (column C in Table 1) were esti- termites were found at these sites. The rates of mated from the distributions of trains rates, burst rates, termite trains and bursts were high at sites where and O. rhinoceros counts in Tables 2Ð4. Rates of larval- termites were found (Table 3). and adult-proÞle trains and bursts below their esti-

Impulse trains that matched the adult or larval O. mated lower cutoff rates in column C of Table 1 seemed rhinoceros proÞles were observed in recordings to be associated with a low likelihood of infestation. where only nontarget organisms were recovered but Rates greater than or equal to their estimated upper generally at lower rates than observed at sites where cutoff rates in Table 1 seemed to be associated with a

O. rhinoceros were recovered (Tables 3 and 4). No high likelihood of infestation. The resultant ratings of rhino_a bursts were detected in recordings from infestation likelihood are listed in Table 5. A statisti-

Table 3. Rates (number per minute) of O. rhinoceros larval-profile trains, rrhino_l-t, and bursts, rrhino_l-b, adult-profile trains, rrhino_a-t, and bursts, rrhino_a-b, and termite-profile trains, rtermite-t, and bursts, rtermite-b, detected in palm trees and logs where only nontarget sound-producing organisms were recovered, listed in order of descending rates of termite bursts, rtermite-b

Larval (r _ ) Adult (r _ ) Termite (r ) rhino l rhino a termite No. invert.a No. vert.b Trains Bursts Trains Bursts Trains Bursts 0.22 0.00 0.00 0.00 29.44 10.62 Ñc Ñ 1.01 0.00 0.00 0.00 25.13 10.05 Ñc Ñ 11.54 0.38 0.00 0.00 41.54 9.23 Ñc Ñ 12.69 0.67 0.00 0.00 36.07 8.68 2 Ñ 5.17 0.40 1.59 0.40 20.67 6.76 4 Ñ 0.89 0.00 25.48 1.79 7.60 3.58 3 2 10.94 1.86 25.81 0.21 9.50 3.30 3 2 31.60 0.72 38.07 0.00 6.46 1.44 5 Ñ 26.85 0.75 38.65 0.00 6.27 0.50 —c Ñ 5.86 0.90 44.63 1.80 1.35 0.45 3 2 37.85 0.49 5.85 0.00 18.93 0.29 Ñc Ñ 3.45 0.00 59.42 0.69 0.00 0.00 3 2 0.00 0.00 40.84 0.00 0.00 0.00 2 Ñ

a B. palauensis, C. morio, P. surinamensis, tenebrionids, diplopodans, with uncounted (Ͼ10) N. luzonicus present where noted (c). b Tree frogs, geckos. c N. luzonicus. August 2010 MANKIN AND MOORE:INVASIVE O. rhinoceros ACOUSTIC DETECTION 1141

Table 5. Distributions of computer-rated likelihood of O. rhi- Table 6. Distributions of computer-rated likelihood of termite noceros infestation at sites where larvae or adults were confirmed infestation at sites where N. luzonicus were confirmed absent or absent or present, by using indicators estimated from observed present, by using indicators estimated from observed distributions distributions of rates of rhino_l or rhino_a impulse trains or bursts of rates of termite impulse trains or bursts

No. sites with O. rhinoceros absent or present, No. sites with N. luzonicus absent or present, Infestation rated by trains or bursts indicator Infestation rated by trains or bursts indicator likelihood likelihood Trainsa Burstsb Trainsa Burstsb rating rating Absent Present Absent Present Absent Present Absent Present

low 13 0 17 0 low 14 1 15 2 medium 61 91medium 31 41 high 72 02high 23 02

a ϭ a ϭ P 0.17 that trains comprehensive indicator, irhino-t, is indepen- P 0.02 that the trains indicator, itermite-t, is independent of dent of absence or presence of O. rhinoceros at recording site (Wil- presence or absence of termites at recording site (Wilcoxon two- coxon two-sample exact test: S ϭ 67, Z ϭ 1.6535, N ϭ 29), where sample exact test: S ϭ 91.5, Z ϭ 2.3495, N ϭ 24), where indicator value Ͻ Ͻ Ͻ Ͼ indicator value is low if (rrhino_l-t 25.0 and rrhino_a-t 15.0), high if is low if rtermite-t 15.0, high if rtermite-t 35.0, and otherwise medium. Ͼ Ͼ b ϭ (rrhino_l-t 50.0 or rrhino_a-t 35.0), and otherwise medium. P 0.03 that bursts indicator, itermite-b, is independent of presence b ϭ P 0.003 that bursts comprehensive indicator, irhino-b is inde- or absence of termites at recording site (Wilcoxon two-sample exact ϭ ϭ ϭ pendent of presence or absence of O. rhinoceros at recording site test: S 85.0, Z 1.9608, N 24), where indicator value is low if ϭ ϭ ϭ Ͻ Ͼ (Wilcoxon two-sample exact test: S 79.5, Z 2.7963, N 29), where rtermite-b 5.0, high if rtermite-b 10.0, and otherwise medium. Ͻ Ͻ indicator value is low if (rrhino_l-b 0.5 and rrhino_a-b 0.5), high if Ͼ Ͼ (rrhino_l-b 5.0 or rrhino_a-b 2.0), and otherwise medium. died without cutting into the tree, two were alive but had caused no damage, and 10 others had burrowed 5 cally signiÞcant relationship was found between the cm or more into the trunk. bursts comprehensive indicator, irhino-b (equation 2, Groups of larvae and adults in rearing boxes pro- and the presence or absence of O. rhinoceros, but the duced audible sounds at rates of 6 to 7 trains per min relationship for the trains comprehensive indicator, in pieces of palm tree trunk and sugarcane. Although irhino-t (equation 1, was not statistically signiÞcant. the spectral patterns were different from those of Distributions of termite Trains and Bursts Relative sounds produced in trees and logs, due to the different to N. luzonicus and O. rhinoceros Infestations. Ter- substrates, both larvae and adults conducted activities mites were encountered at Þve recording sites, and we with impulse rate patterns similar to those in the nat- considered whether the rates of termite trains, rtermite-t, ural infestations. In the boxes with feeding larvae, the and bursts, rtermite-b, could be used to construct indi- larvae typically oriented head-to-tail inside an approx- cators of termite infestation likelihood similar to the imately spherical chamber, which they had burrowed indicators of O. rhinoceros infestation. The distribu- out of the palm trunk. They moved in a circular or tions of burst rates, trains rates, and N. luzonicus counts toroidal pattern, with a moving constriction and then in Tables 2 and 3 were used to estimate lower cutoff an expansion that began at the tip of the abdomen and rates, below which N. luzonicus were rarely present, moved toward the head. The movement produced an and upper cutoff rates, above which N. luzonicus were audible sound, and the timing of a typical burst cor- likely to be found (Table 1). It should be noted that responded to the timing of the constrictionÐexpansion the termite analysis did not include results from the pattern. Þve stumps in Table 4 because it was not possible to determine conclusively that termites were present or Discussion absent in the unexcavated portions of the root system. Computer ratings of infestation likelihood based on The notable audibility of sounds produced by O. the above cutoff rate estimates are listed in Table 6. rhinoceros adults and larvae has attracted the interest Both trains and bursts indicators were found to be of several generations of entomologists (Darwin 1871, signiÞcantly associated with the presence of N. luzoni- Gressitt 1953, Mini and Prabhu 1990). Thus, it is no cus at the recording sites. It was thus possible in this surprise that adult O. rhinoceros scraping and chewing study to predict independently the presence or ab- movements and larval burrowing activities produce sence of both O. rhinoceros and termites at a recording sounds with distinctive spectral and temporal patterns site using acoustic indicators. that facilitate detection by acoustic instruments in Signals From O. rhinoceros Confined in Cages or urban environments. Other examples have been re- Rearing Boxes. To obtain conÞrmatory recordings of ported of leafminer larvae (Bacher et al. 1996, Mey- adult-produced sounds, 14 adults collected from pher- ho¨fer and Casas 1999, Castellanos and Barbosa 2006, omone-baited traps were placed in separate cages that Casas and Magal 2007, Low 2008) and termites (Inta were attached to palm trees with an open side per- et al. 2009) that perform evasive or avoidance behav- mitting entry to the trunk. Recordings were collected iors upon detecting distinctive spectral and temporal on the following day, and the cages were subsequently patterns in sounds produced by parasitoids and pred- removed. Adult-proÞle trains were detected in the ators. Distinctive frequency and temporal patterns recordings from all of the artiÞcially infested palm of predator vibrations are known to elicit escape trees. However, adult-proÞle bursts were detected at hatching of red-eyed tree frogs (Caldwell et al. only one tree. Upon inspection, two of the adults had 2009). The sounds of burrowing moles have been 1142 JOURNAL OF ECONOMIC ENTOMOLOGY Vol. 103, no. 4 shown to elicit escape activities in earthworms and References Cited have been mimicked by worm grunters who har- “ ” Acda, M. N. 2007. Toxicity of thiamethoxam against Philip- vest earthworms for Þshing (Catania 2008, Mitra et pine subterranean termites. J. Insect Sci. 7: 26. al. 2009). To survive, of many different Bacher, S. J. Casas, and S. Dorn. 1996. Parasitoid vibrations species routinely detect and identify distinctive as potential releasing stimulus of evasive behaviour in a spectral and temporal patterns of acoustic signals leafminer. Physiol. Entomol. 21: 33Ð43. produced by other organisms, and such capability Bedford, G. O. 1980. Biology, ecology, and control of palm was perfected many thousands of years before the rhinoceros beetles. Annu. Rev. Entomol. 25: 309Ð339. development of equivalent digital signal processing Caldwell, M. S., J. G. McDaniel, and K. M. Warkentin. 2009. Frequency information in the vibration-cued escape methods that are only now are coming into use. hatching of red-eyed treefrogs. J. Exp. Biol. 212: 566Ð575. It is likely, however, that the reliability of O. rhi- Casas, J., and C. Magal. 2007. Mutual eavesdropping through noceros detection could be enhanced in the future vibrations in a host-parasitoid interaction: from plant bi- beyond what was accomplished in this study. Larvae omechanics to behavioural ecology, pp. 263Ð271. In S. and adults (Mankin et al. 2009) both produce stridu- Drosopoulos and M. F. Claridge [eds.], Insect sounds and lations, for example, and more detailed characteriza- communication: physiology, behaviour, ecology and evo- lution. CRC, Boca Raton, FL. tion of stridulations and scraping motions could lead Castellanos, I., and P. Barbosa. 2006. Evaluation of preda- to further increases in reliability of detection. In ad- tion risk by a caterpillar using substrate-borne vibrations. dition, the short trains underlined in Figs. 1 and 2 Anim. Behav. 72: 461Ð469. possibly were produced by O. rhinoceros, but did not Catania, K. C. 2008. Worm grunting, Þddling, and charming. ϭ meet the criterion of nmin-rhino_a and nmin-rhino_l 20 Humans unknowingly mimic a predator to harvest bait. impulses per train, equivalent to a relatively long, PLoS One 3: e3472. 60Ð200-ms minimum duration, adopted to reduce Charif, R. A., A. M. Waack, and L. M. Strickman. 2008. Raven Pro 1.3 userÕs manual. Cornell Laboratory of Or- false-positive identiÞcations of background noise or nithology, Ithaca, NY. nontarget insects as O. rhinoceros sounds. IdentiÞca- Darwin, C. 1871. The ascent of man, and selection in rela- tion of multiple insect-sound patterns that are unlikely tion to sex. vol. 1. John Murray, London, United Kingdom. to occur randomly in background sounds can lead not Fukuda, K., S. Utsuzawa, and D. Sakaue. 2007. Correlation only to improved detection methods but also to im- between acoustic emission, water status and xylem em- proved understanding of insect behavior. bolism in pine wilt disease. Tree Physiol. 27: 969Ð976. The success of the detection tests conducted in this Gressitt, J. L. 1953. The coconut rhinoceros (Oryctes rhinoceros) with particular reference to the Palau Islands. study suggests that acoustic methods could be incor- Bulletin 212, Bishop Museum, Honolulu, HI. porated beneÞcially into the ongoing program to erad- Hallett, R. H., A. L. Perez, G. Gries, R. Gries, H. D. Pierce, icate the introduced O. rhinoceros infestations from Jr., J. Yue, A. C. Oehslchlager, L. M. Gonzalez, and J. H. Guam. The O. rhinoceros signals are distinctive, which Borden. 1995. Aggregation pheromone of coconut rhi- facilitates their identiÞcation by scouts or by subse- noceros beetle, Oryctes rhinoceros (L.) (Coleoptera: quent computer analyses of recorded signals. This Scarabaeidae). J. Chem. Ecol. 21: 1549Ð1570. helps enable detection of active adults in crowns of Hinckley, A. D. 1973. Ecology of the coconut rhinoceros beetle, Oryctes rhinoceros (L.), (Coleoptera: Dynasti- live trees and larvae in standing dead trunks that oth- dae). Biotropica 5: 111Ð116. erwise might escape detection. In each survey loca- Inta, R., T. A. Evans, and J.C.S. Lai. 2009. Effect of vibratory tion, the scout can decide separately at each tree soldier alarm signals on the foraging behavior of subter- whether a visual inspection is sufÞcient to determine ranean termites (Isoptera: Rhinotermitidae). J. Econ. En- the presence or absence of infestation, or if time is tomol. 102: 121Ð126. available to attach an acoustic sensor to the trunk and Jackson, T. A., and M. G. Klein. 2006. Scarabs as pests: a continuing problem. Coleopt. Bull. 60: 102Ð119. listen for distinctive sounds produced by O. rhinoceros. Low, C. 2008. Seismic behaviors of a leafminer, Antispila The additional time needed for recording is 5Ð15 min nysaefoliella (Lepidoptera: Heliozelidae). Fla. Entomol. per tree, which often is less than the time needed to 91: 604 609. move between two separate survey locations. Mankin, R. W. 1994. Acoustical detection of Aedes taenio- rhynchus swarms and emergence exoduses in remote salt marshes. J. Am. Mosq. Control Assoc. 10: 302Ð308. Mankin, R. W., and J. Benshemesh. 2006. Geophone detec- Acknowledgments tion of subterranean termite and ant activity. J. Econ. Entomol. 99: 244Ð250. Robert Bourgeois (University of Guam), Roland Quitugua Mankin, R. W., J. Brandhorst-Hubbard, K. L. Flanders, M. (Guam Coconut Rhinoceros Beetle Eradication Project), Zhang, R. L. Crocker, S. L. Lapointe, C. W. McCoy, J. R. Everett Foreman (ARS), and Betty Weaver (ARS) provided Fisher, and D. K. Weaver. 2000. Eavesdropping on in- technical support and also participated at different times as sects hidden in soil and interior structures of plants. J. listeners on site or during subsequent playbacks of recordings Econ. Entomol. 93: 1173Ð1182. in Raven along with R.W.M. and A.M. Jan Krecek and Rudolf Mankin, R. W., J. L. Hubbard, and K. L. Flanders. 2007. Scheffrahn (Ft. Lauderdale Research and Education Center, Acoustic indicators for mapping infestation probabilities Institute of Food and Agricultural Sciences, University of of soil invertebrates. J. Econ. Entomol. 100: 790Ð800. Florida) provided identiÞcations of termite specimens. Fi- Mankin, R. W., S. L. Lapointe, and R. A. Franqui. 2001. nancial support was provided in part by a Western Integrated Acoustic surveying of subterranean insect populations in Pest Management Center Special Issues Grant. citrus groves. J. Econ. Entomol. 94: 853Ð859. August 2010 MANKIN AND MOORE:INVASIVE O. rhinoceros ACOUSTIC DETECTION 1143

Mankin, R. W., A. Mizrach, A. Hetzroni, S. Levsky, Y. Na- Ramle, M., M. B. Wahid, K. Norman, T. R. Glare, and T. A. kache, and V. Soroker. 2008a. Temporal and spectral Jackson. 2005. The incidence and use of Oryctes virus for features of sounds of wood-boring beetle larvae: identi- control of rhinoceros beetle in oil palm plantations in Þable patterns of activity enable improved discrimination Malaysia. J. Invert. Pathol. 89: 85Ð90. from background noise. Fla. Entomol. 91: 241Ð248. SAS Institute. 2004. SAS/STAT 9.1 userÕs guide. SAS Insti- Mankin, R. W., A. Moore, P. R. Samson, and K. J. Chandler. tute, Cary, NC. 2009. Acoustic characteristics of dynastid beetle stridu- Smith, S. L., and A. Moore. 2008. Early detection pest risk as- lations. Fla. Entomol. 92: 123Ð134. sessment: coconut rhinoceros beetle. (http://guaminsects. Mankin, R. W., W. L. Osbrink, F. M. Oi, and J. B. Anderson. net/uogces/kbwiki/images/1/13/CRB_Pest_Risk_ 2002. Acoustic detection of termite infestations in urban Assessment.pdf). trees. J. Econ. Entomol. 95: 981Ð988. Snyder, T. E., and F. C. Francia. 1960. A summary of Phil- Mankin, R. W., M. T. Smith, J. M. Tropp, E. B. Atkinson, and ippine termites with supplementary biological notes. D. Y. Jong. 2008b. Detection of Anoplophora glabripen- Philipp. J. Sci. 89: 63Ð77. nis (Coleoptera: Cerambycidae) larvae in different host Su, N.-Y., and R. H. Scheffrahn. 1998. Coptotermes vastator trees and tissues by automated analyses of sound-impulse Light (Isoptera: Rhinotermitidae) in Guam. Proc. Hawai- frequency and temporal patterns. J. Econ. Entomol. 101: ian Entomol. Soc. 33: 13Ð18. 836Ð849. Zhang, M., R. L. Crocker, R. W. Mankin, K. L. Flanders, Meyhofer, R., and J. Casas. 1999. Vibratory stimuli in host ¨ and J. L. Brandhorst-Hubbard. 2003a. Acoustic iden- location by parasitic wasps. J. Insect Physiol. 45: 967Ð971. tiÞcation and measurement of activity patterns of white Mini, A., and V.K.K. Prabhu. 1990. Stridulation in the co- grubs in soil. J. Econ. Entomol. 96: 1704Ð1710. conut rhinoceros beetle Oryctes rhinoceros (Coleoptera: Zhang, M., R. L. Crocker, R. W. Mankin, K. L. Flanders, and Scarabaeidae). Proc. Indian Acad. Sci., Anim. Sci. 99: 447Ð455. J. L. Brandhorst-Hubbard. 2003b. Acoustic estimation Mitra, O., M. A. Callaham, Jr., M. L. Smith, and J. E. Yack. of infestations and population densities of white grubs 2009. Grunting for worms: seismic vibrations cause Dip- (Coleoptera: Scarabaeidae) in turfgrass. J. Econ. Ento- locardia earthworms to emerge from the soil. Biol. Lett. mol. 96: 1770Ð1779. 5: 16Ð19. Muniappan, R. 2002. Pests of coconut and their natural enemies in Micronesia. Micronesica Suppl. 6: 105Ð110. Received 28 June 2009; accepted 21 November 2009.