Maize is an important source of income and caloric intake in sub-Saharan Africa, accounting for almost 20% of the plant-based food supply (Jones et al. 2011). However, the maize crop can experience up to 20–30% loss yearly because farmers often lack effective postharvest storage technologies (Mwololo et al. 2010). In rural regions where maize typically is stored in bags or other containers (De Groote et al. 2013; Abass et al. 2014), these losses are due primarily to 2 internally feeding post-harvest insect pests, Prostephanus truncatus Horn (Coleoptera: Bostrichidae) and Sitophilus zeamais Motschulsky (Coleoptera: Curculionidae) (Vowotor et al. 2005; Omondi et al. 2009; Tefera et al. 2011). In Tanzania, indigenous storage facilities include the traditional kihenge, a raised grass-thatched, mud-plastered hut on stilts (Rugumamu 2003; Makalle 2012). Efforts to detect insects in kihenge and other maize containers typically include frequently smelling and visually inspecting the maize for indications of spoilage, as well as long-term monitoring of its temperature, all of which are time-consuming and costly (Anonymous 2011; Abass et al. 2014). Because the larvae feed unseen inside the kernels, infestations often remain undetected until considerable damage has occurred. Consequently, there is need for nondestructive instruments that can help guide the efforts of pest managers by inexpensively and rapidly detecting infestations in traditional, small-scale storage facilities (Rugumamu 2003).

Acoustic methods are available for detecting infestations of internally feeding insects, but commercially marketed instruments (Litzkow et al. 1990; Hagstrum et al. 1996; Neethirajan et al. 2007; Mankin et al. 2011; Rohde et al. 2013) have been too expensive for general use in small-scale storage facilities in developing countries. However, recent innovations and improvements in the usability of electronic sensing systems suggest it may be possible to develop a low-cost acoustic detection device (e.g., Mankin et al. 2010, 2013; Dillman et al. 2014) incorporating Arduino, Raspberry Pi, or other microcontrollers that could be adopted more widely than previous devices as a tool for insect pest management in maize- and other grain-storage facilities in sub-Saharan Africa.

To distinguish the sounds of a targeted insect species from background noise, such a device amplifies and records incoming sounds and then matches the spectra of individually detected sound impulses to amplitudes and mean spectra (profiles) of sound impulses verified to be produced by that species (Mankin et al. 2011). Some of the first studies of the spectra of sounds produced by internally feeding stored product insects were conducted by Vick et al. (1988b) on Sitophilus oryzae (L.), Rhyzopertha dominica (F.), and Sitotroga cerealella (Olivier) in maize, wheat, and rice kernels. No consistent differences were observed in the spectral ranges of sounds produced by the 3 insects. The peak signal amplitudes were at 587, 1,200, and 1,475 Hz, for larvae feeding in maize, wheat, and rice, respectively. These differences in peak frequency are consistent with differences in the sizes of the grain kernels, in that the resonant frequencies of structural fibers in larger kernels would be expected to be lower than the resonant frequencies of structural fibers in smaller kernels (Mankin et al. 2011).

The stored product insect study of Vick et al. (1988b) did not characterize the full spectral range of sound impulses produced by internally feeding larvae, which extends broadly into ultrasonic frequencies (Pittendrigh et al. 1997; Mankin 2011). To develop a more complete set of larval sound profiles that would be useful for matching to signals collected by an electronic device in a storage facility, we conducted a study of sounds produced in maize by S. oryzae larvae. To determine effects of distance from sensor on the detectability of larval sounds, we examined signals at several different distances between the larvae and the sensor. Sitophilus oryzae is a widely distributed stored product insect pest whose larvae are slightly smaller than those of S. zeamais, but the 2 species are similar in their feeding behavior (Alleoni & Ferreira 2006).

Materials and Methods

Sitophilus oryzae larvae were obtained from a colony at the United States Department of Agriculture (USDA) Agricultural Research Service (ARS) Center for Medical, Agricultural, and Veterinary Entomology (CMAVE), Gainesville, Florida, USA. Infested wheat kernels from the colony (25 mL) were put in a 60 mL plastic pouch, which then was placed near the bottom or near the outer edge of a 27.7 kg polyethylene bag of cracked corn during testing. The number and ages of larvae that produced sound during each recording were undetermined, but we estimated that multiple 4th instars were feeding during each test, given that 83 adult S. oryzae had emerged within the pouch 10 d after the experiment was completed. The typical pattern of S. oryzae development in grain kernels is: first instar at day 5 after egg is laid, 2nd at day 9, 3rd at day 13, 4th at day 16, and adult emergence at day 32 (Howe 1952).

To record signals at different distances between larvae and sensors, a 15-cm-length waveguide (probe) was attached to a sensor/preamplifier (Model SP-1L, Acoustic Emission Consulting [AEC] Inc., Sacramento, California, USA) of an acoustic insect detection system (Model AED-2010, AEC. Inc.). The probe was inserted into the bag of corn at different horizontal or vertical distances 5–35 cm from the nearest edge of the pouch. The tests were conducted at 24–27 °C in a sound- and vibration-shielded anechoic chamber that reduced background noise to negligible levels (Vick et al. 1988a; Mankin et al. 1996). Three recordings conducted on different days without the pouch in the bag of corn confirmed that background sounds occurred at a rate of < 0.01 s^-1.

Intervals of 180 s that contained 3- to 30-ms sound impulses (Vick et al. 1988b; Mankin et al. 1996, 2011) were amplified 78 dB (Model AED-2010, AEC, Inc.), where dB is 20 log₁₀(V_output) and V_output is the signal delivered by the amplifier in units of microvolts. The amplified signals were collected at a 44.1 kHz sampling rate onto a digital audio recorder (Model PMD661, Marantz, Mahwah, New Jersey, USA). General-purpose sound analysis software (Raven Pro Version 1.5; Charif et al. 2008) was used to screen and preprocess the recordings. Spectral and temporal pattern analyses of the larval impulses were conducted using DAVIS, a custom-written insect sound analysis program (Herrick & Mankin 2012; Dosunmu et al. 2014). It should be noted that the AED-2010, whose frequency bandwidth extends to 2 MHz, filters out signals below 1 kHz to reduce the level of background noise.

Fig. 1.

Spectral profiles of 4 distinctive types of larval sound impulses detected in cracked corn: HaNb, solid line; Ma, dashed line, Ha, dash-dot-dotted line, and La, dotted line. Horizontal axis indicates frequency in kHz and vertical axis indicates relative spectrum amplitude in dB.

Results

Four distinct types of impulses were identified in the initial screening. Spectral profiles of each type (Mankin et al. 2011) were generated in DAVIS by calculating the mean spectrum of a series of consecutive impulses of the same type observed in a single recording. Each impulse detected in the recordings was assigned to the type from which it had the smallest total mean-square difference (Dosunmu et al. 2014). About 1% of the detected signals had a total least-squares difference greater than 50 dB against all of the profiles, and these were discarded as background noise. For designative purposes, the 4 signal types were labeled: HaNb (High amplitude-Narrow band), Ma (Mid-range amplitude), Ha (High amplitude), and La (Low amplitude). The spectral profiles of the S. oryzae larval impulse types are shown in Fig. 1 in order of lowest to highest occurrence of impulse matches. Only 39 of 2,741 larval impulses total were best-fit matched to the HaNb profile, and impulses of this type were found in only 3 of the 11 tests where larval impulses were detected. However, these impulses were very prominent whenever they occurred, because HaNb had the highest amplitude of all the profiles, primarily in a narrow band of frequencies around 6.2 kHz. The HaNb profile was generated as an average of 7 consecutive impulses in a recording of signals 10 cm from the larval pouch.

Fig. 2.

Oscillogram of sound impulses recorded 10 cm from pouch containing Sitophilus oryzae larvae. Examples of 3 types of larval sound impulse occur during the 1 s period, and one example each of type (Ha, La, and HaNb) is marked above the impulse. Horizontal axis indicates time in seconds and vertical axis indicates relative signal amplitude.

There were 196 impulses that matched the mid-range amplitude profile, type Ma, which was generated as an average of 97 consecutive impulses in a recording at 10 cm from the larval pouch. Ten of the 11 tests with larval impulses contained examples of Ma impulses, and all of them contained examples that matched the type Ha and type La profiles. The Ha profile was generated as an average of 41 consecutive impulses in a recording of signals at 5 cm distance from the larval pouch, and the La profile was generated from an average of 8 consecutive impulses in a test at a 20 cm distance from the larval pouch. A total of 1,132 impulses of type Ha were detected, and 1,374 of type La. Figure 2 shows a 1 s interval recorded 10 cm away from the larval pouch and includes 3 of the 4 types of sound impulses. Six impulses of type Ha precede the type Ha impulse marked at 0.37 s. A pair of type La impulses is marked at 0.52 s, and 2 type HaNb impulses appear at the end of the interval, near 0.97 s.

For mean and standard error calculations to consider the effects of distance between the larval pouch and sensor on the detectability of larvae in maize, the rates of larval sound impulses of each profile type were summed and averaged over the test duration to obtain an overall rate of impulses for each recording. Because the variability of rates was similar for recordings collected at the same horizontal and vertical distances from the pouch, overall rates were combined into 4 groupings of mean distance: 8, 18, 27, and 35 cm, with n = 4, 4, 3, and 3 tests, respectively. The results are shown in Fig. 3. The larvae were easily detected at mean distances of 8 cm but were undetectable at distances of 35 cm. In general, the effect of distance was consistent with the conclusion of Vick et al. (1988a) that the rate of sounds decreased approximately logarithmically with distance between the larvae and the sensor. The pattern of detectable sounds in Fig. 3 suggests that, except for very loud larvae, the maximum range of detection is no greater than about 30 cm in a quiet environment.

Fig. 3.

Effects of distance on detectability of larval sound impulses. Horizontal axis indicates the mean distance between the larval pouch and the sensor; vertical axis indicates the log₁₀-transformed mean rate of impulses detected at that distance. Bars indicate the standard error of mean transformed rate.

Discussion

The spectral ranges of impulses detected in this study are similar to those observed in other studies of insect sounds produced in stored maize or wheat (Vick et al. 1988b; Mankin et al. 2011). Therefore, it can be expected that any insect acoustic detection instrument designed for small-scale storage facilities would need to detect signals in the range of 500–7,000 Hz to ensure coverage of the most prominent frequencies of signals produced by internally feeding stored product insects in maize. Fortunately, such capabilities could be incorporated into an inexpensive computer system, such as an Arduino microcontroller ( http://arduino.cc/en/Main/ArduinoBoardUno) or a Raspberry Pi platform ( http://www.raspberrypi.org), which can be used with open-source or custom-written data collection and analysis software and inexpensive memory storage cards. Such systems offer the added advantage that they can be battery powered, as many parts of sub-Saharan Africa do not have reliable electrical power; it would also be beneficial to include a feature that enables the device to communicate with increasingly available mobile phones (Aker & Mbiti 2010). Because it is unlikely that any low-cost insect acoustic detection instrument would have greater sensitivity than the state-of-art system used in this study, it can be expected that P. truncatus and S. zeamais would be detected by any instrument only rarely at distances greater than 30 cm from the sensor. However, probes of different sizes could be developed to adapt the sensors for use with the different types of storage containers used in the region (e.g., polyethylene bags, jute bags, and kihenge). Thus, if an inexpensive, robust sensor can be developed and operated with a low-cost microcontroller system, there is considerable potential for its use in insect pest management activities in rural Tanzania and other regions of sub-Saharan Africa.