A wide range of imaging and spectroscopy technologies is used in medical diagnostics, quality control in production systems, military applications, stress detection in agriculture, and ecological studies of both terrestrial and aquatic organisms. In this study, we hypothesized that reflectance profiling can be used to successfully classify animals that are otherwise very challenging to classify. We acquired hyperspectral images from adult specimens of the egg parasitoid genus Trichogramma (T. galloi, T. pretiosum and T. atopovirilia), which are ~1.0 mm in length. We also acquired hyperspectral images from host eggs containing developing Trichogramma instar and pupae. These obligate egg endoparasitoid species are commercially available as natural enemies of lepidopteran pests in food production systems. Because of their minute size and physical resemblance, classification is time consuming and requires a high level of technical experience. The classification of reflectance profiles was based on a combination of average reflectance and variogram parameters (describing the spatial structure of reflectance data) of reflectance values in individual spectral bands. Although variogram parameters (variogram analysis) are commonly used in large-scale spatial research (i.e. geoscience and landscape ecology), they have only recently been used in classification of high-resolution hyperspectral imaging data. The classification model of parasitized host eggs was equally successful for each of the three species and was successfully validated with independent data sets (>90% classification accuracy). The classification model of adult specimens accurately separated T. atopovirilia from the other two species, but specimens of T. galloi and T. pretiosum could not be accurately separated. Interestingly, molecular-based classification (using the DNA sequence of the internally transcribed spacer ITS2) of Trichogramma species published elsewhere corroborates the classification, as T. galloi and T. pretiosum are closely related and comparatively distant from T. atopovirilia. Our results emphasize the importance of using high-spectral and high-spatial resolution data in the classification of organism relatedness, and hyperspectral imaging may be of relevance to a wide range of commercial (i.e. producers of biocontrol agents), taxonomic and evolutionary research applications.

INTRODUCTION

A recent review article claims that UAVs (unmanned aerial vehicles) will revolutionize research methods in spatial ecology (Anderson and Gaston, 2013). Progressive expansion of image-based applications in both research and commercial applications is driven by a series of trends, including: progressive improvements in the robustness of vehicles and sensors, lighter and more powerful technologies, lower cost of equipment, and advances in processing and classification of imaging data. Here, we argue that advances in the development of imaging technologies not only improve our ability to acquire, process and interpret spatial data on a large geographical scale but also enable new methods of classifying organisms (animals and growing plants and seeds) and non-destructively assessing their relatedness.

The intensity and composition of photons reflected (here referred to as reflectance profiles) from an object is partially determined by the radiometric energy source (lighting), projection angle, environmental conditions (i.e. temperature and barometric pressure), and the physical structure and/or biochemical composition of the objects' surface (Clark, 1999). Thus, if the lighting source, projection angle and environmental conditions are maintained constant, distinct differences in reflectance profiles between target objects are assumed to be associated with differences in the physical structure and/or biochemical composition of the target objects. Hyperspectral imaging technologies have been used to provide insight into a wide range of biological phenomena, including: (1) how color-blind cuttlefish [Sepia officinalis L. (Cep.: Sepiida)] camouflage themselves (Chiao et al., 2011), (2) twilight spectral dynamics and their effects on mass spawning events on coral reefs (Sweeney et al., 2011), (3) vision in honeybees [Apis mellifera L. (Hym.: Apidae)] and orb-web spiders [Nephila pilipes Fabricius (Ara.: Nephilidae)] (Chiao et al., 2009) and (4) the role of an enlarged claw in mating courtship and territorial displays among fiddler crabs [Uca vomeris McNeill (Dec.: Ocypodidae)] (Zeil and Hofmann, 2001). In addition, reflectance-based technologies have been used to differentiate: species of stored grain insects (Singh et al., 2010), tobacco budworm [(Heliothis virescens (Fabricius) (Lep.: Noctuidae)] and corn earworm [(Helicoverpa zea (Boddie) (Lep.: Noctuidae)] (Jia et al., 2007), and to age-grade biting midges [Culicoides sonorensis Wirth and Jones (Dip.: Ceratopogonidae)] (Reeves et al., 2010). Klarica et al. (Klarica et al., 2011) used imaging spectroscopy to discriminate cryptic species of ants [Tetramorium caespitum (L.) and T. impurum (Foerster) (Hym.: Formicidae)]. Regarding classification based on reflectance profiles acquired from the external surface of insects and other organisms, both physical and biochemical factors are likely involved. It was beyond the scope of this study to identify and characterize the possible mechanisms responsible for species-specific differences in reflectance profiles, but the density and size of body hairs, body shape, and ornamentation and pigmentation of the insect cuticle are believed to be of major importance. As an example, the hydrocarbon composition of insect cuticle is known to be highly dynamic and vary: (1) among closely related species (Dowell et al., 1999; Roux et al., 2008), (2) in relative composition (Geiselhardt et al., 2009) or in actual composition (Howard and Pérez-Lachaud, 2002) among males and females within a species, (3) during the lifespan of insect individuals (Butler et al., 2009; Howard and Baker, 2003; Lapointe et al., 2004; Roux et al., 2008; Zhu et al., 2006), (4) among eusocial individuals with different tasks (Ferreira-Caliman et al., 2010; Nunes et al., 2009), (5) according to mating behavior and status (Howard and Baker, 2003; Howard and Pérez-Lachaud, 2002; Steiner et al., 2007), and (6) in response to environmental conditions (Howard and Baker, 2003; Howard and Pérez-Lachaud, 2002; De Loof et al., 2010). Thus, there is ample evidence of the composition of the insect cuticula being highly dynamic and tightly associated with complex internal physiological processes in insects. And differences in the biochemical composition of the cuticula would imply corresponding differences in reflectance profiles and therefore the possibility of classifying even closely related species.

Trichogramma species are obligate egg endoparasitoids; about 210 Trichogramma species have been identified (Pinto, 2006) and they are among the mostly commercialized natural enemies for control of Lepidopteran pests (Li, 1994). Commercial mass-rearing of Trichogramma species occurs in at least 16 countries, including: (1) Russia, (2) Mexico, (3) China (Lenteren, 2008), (4) Cuba and (5) Brazil (Parra, 2010). Trichogramma species are released into a wide range of cropping systems, including horticultural crops, cereals, cotton (Gossypium spp. L.), corn (Zea mays L.), sugarcane (Saccharum officinarum L.), tobacco (Nicotiana tabacum L.) and pastures (Lenteren, 2008; Li, 1994; Parra, 2010). The success of Trichogramma spp. as biocontrol agents is partially linked to the possibility of mass rearing them on factitious host eggs from stored grain moth species, such as the Mediterranean flour moth [Ephestia kuehniella Zeller (Lep.: Pyralidae)] (Hassan, 1997; Lewis, 1976; Parra, 2010). However, Danks (Danks, 1988) and Smith (Smith, 1996) reported on several examples of costly biocontrol releases with limited effect on target pest populations, and the lack of success was partially attributed to misclassification and/or cross-contamination of species during the mass-rearing process. Thus, numerous studies have investigated the variability in quality of mass-reared Trichogramma species (Bigler, 1989; Cerutti and Bigler, 1995; Kuhlmann and Mills, 1999; Losey and Calvin, 1995; Lundgren and Heimpel, 2003). Zucchi and Querino (Zucchi and Querino, 2010) argued that continuous taxonomic monitoring of Trichogramma populations is important both in commercial mass-rearing facilities and under field conditions after mass releases. However, species identification of micro-hymenoptera requires a high level of technical experience and is also time consuming (Pinto and Stouthamer, 1994). Furthermore, as discussed by Almeida (Almeida, 2004), identification of Trichogramma species has in the past been widely based on male genitalic characteristics, which is a particular challenge as many Trichogramma species have partial or completely parthenogenic forms. Furthermore, identification based on male genitalic characteristics involves considerable practical experience, and it may also require fairly elaborate chemical procedures, including chemical treatments and correct mounting of genitalia on microscope slides (Querino and Zucchi, 2011). As part of a drive to improve methods used to classify Trichogramma species, Almeida (Almeida, 2004) discussed the potential of a range of molecular-based methods and developed a classification key for 17 South American species based on the DNA sequence of the internally transcribed spacer ITS2 (Stouthamer et al., 1999). The classification key was based on the size of the nuclear ribosomal gene ITS2 PCR product and three endonucleases (EcoRI, MseI and MaeI). Recently, Querino and Zucchi (Querino and Zucchi, 2011) published a new key to Trichogramma species, but it is still important to emphasize that identification of minute adult Trichogramma specimens (1.0 mm in length) is challenging and requires considerable technical experience.

In this study, we hypothesized that classification of closely related animals can be conducted successfully on the basis of the spatial structure of high spatial (45 pixels mm−2) and spectral (160 spectral bands from 405–907 nm) resolution imaging data. As a challenging model system of reflectance-based classification of insects and possibly other living organisms (other animals, growing plants and seeds), we acquired hyperspectral images of three Trichogramma species, which are obligate egg endoparasitoids. We acquired hyperspectral images of adult specimens (T. galloi Zucchi; T. pretiosum Riley and T. atopovirilia Oatman and Platner) and of parasitized host eggs. The classification of reflectance profiles was based on a combination of average reflectance and variogram parameters (describing the spatial structure of reflectance data) of reflectance values in individual spectral bands. Our data suggest that classification of hyperspectral imaging data may be used both to identify species and also to characterize their relatedness.

RESULTS

Identification of parasitized host eggs

Fig. 1A shows representative photos of non-parasitized (grayish) and parasitized (dark) host eggs, and Fig. 1C clearly indicates that parasitism of host eggs was detectable for the entire spectral range between 405 and 907 nm. Fig. 1B shows representative photos of host eggs parasitized by the three Trichogramma species, and it can be seen that they were virtually indistinguishable based on visual inspection. However, average reflectance profiles in Fig. 1Csuggested that eggs parasitized by T. pretiosum were consistently darker than those parasitized by T. atopovirilia or T. galloi in all spectral bands beyond 600 nm. Thus, based on qualitative/visual assessment of average reflectance profiles of parasitized host eggs (Fig. 1C), there was substantial justification for pursuing a classification approach in which reflectance data in selected spectral bands were used to differentiate host eggs parasitized by different Trichogramma species. A total of 6500 pixels (mean ± s.e. of 108.13±4.45 pixels per host egg) were included after deploying a radiometric filter (Eqn 1) to hyperspectral images of parasitized host eggs. Based on derived variogram parameters (Eqn 3 and Fig. 2), we found that the spectral structure (variogram regression fit) of reflectance data at 687 nm (Fig. 2A) could be used to accurately separate host eggs parasitized by the three Trichogramma species with 86.67% accuracy (Table 1). All three Trichogramma species were classified with similar accuracy. Fig. 2B shows mean reflectance at 687 nm, which also varied significantly among Trichogramma species (d.f.=2,59, F=27.76, P<0.001), but there was no significant difference between T. atopovirilia and T. galloi. Thus, simple comparisons of reflectance averages at 687 nm could not be considered a reliable indicator of all three classes of parasitized host eggs. The linear discriminant classification model based on variogram parameters was validated with reflectance data from 25 independent host eggs parasitized by T. atopivirilia or T. pretiosum, which were classified with 91.17% accuracy (Table 1).

Fig. 1.

Host eggs parasitized by Trichogramma. (A) Images of host eggs with (dark) or without (grayish) parasitism. (B) Host eggs parasitized by (1) T. atopovirilia, (2) T. galloi and (3) T. pretiosum. (C) Relative reflectance from host eggs in 160 spectral bands from 405 to 905 nm.

Fig. 1.

Host eggs parasitized by Trichogramma. (A) Images of host eggs with (dark) or without (grayish) parasitism. (B) Host eggs parasitized by (1) T. atopovirilia, (2) T. galloi and (3) T. pretiosum. (C) Relative reflectance from host eggs in 160 spectral bands from 405 to 905 nm.

Fig. 2.

Classification of parasitized host eggs. (A) Mean c parameter (derived from Eqn 3 and (B) mean (±s.e.) reflectance data at 687 nm of parasitized host eggs. Different letters represent a significant difference at the 0.05 level.

Fig. 2.

Classification of parasitized host eggs. (A) Mean c parameter (derived from Eqn 3 and (B) mean (±s.e.) reflectance data at 687 nm of parasitized host eggs. Different letters represent a significant difference at the 0.05 level.

Identification of Trichogramma adults

Fig. 3A shows representative Trichogramma adults. The micro-hymenoptera are about 1 mm in length and are virtually indistinguishable based on visual inspection. However, average reflectance profiles from the six classes (three species and males/females) suggested that beyond 700 nm, males and females of T. atopivirilia were slightly darker than reflectance profiles from the other two species (Fig. 3B). However, there was no spectral region with a clear distinction of reflectance profiles between T. pretiosum and T. galloi. Thus, separation of adult specimens based on differences in average reflectance in one or more spectral bands appeared to be considerably more challenging than separation of parasitized host eggs. In total, 6318 pixels were included in the classification of Trichogramma adults with no significant difference in the number of pixels among species (d.f.=2,119, F=0.48, P=0.620) but with males (55.93±2.28 pixels) being significantly larger (more pixels acquired per specimen) than females (49.37±1.56 pixels) (d.f.=1,119, F=5.57, P=0.019). Testing variogram parameters in individual spectral bands in 10-band intervals between 405 and 907 nm, we found that a combination of variogram parameters and average reflectance acquired from six spectral bands (465, 497, 529, 560, 718 and 813 nm) could be used to accurately classify adult specimens of the three Trichogramma species with 85.0% accuracy (Table 2). Using only average reflectance in these six spectral bands (not using variogram parameters) enabled classification of adult Trichogramma specimens with 74.17% accuracy. In other words, including variogram parameters in the linear discriminant analysis added considerable accuracy to the classification. The linear discriminant classification model of adult Trichogramma specimens was validated with reflectance data from nine hyperspectral images of 10 (T. galloi only) or 12 adult (10 T. galloi and two T. atopivirilia or T. pretiosum) Trichogramma specimens. In validation analyses of hyperspectral images of 10 T. galloi specimens, none of the specimens were identified as T. atopivirilia, but 53% of them were incorrectly classified as T. pretiosum. In validation analyses of hyperspectral images of 10 T. galloi specimens mixed with two T. pretiosum specimens, none of the specimens were identified as T. atopivirilia, but 32% of them were incorrectly classified as T. pretiosum (higher proportion than two out of 12 individuals). In validation analyses of hyperspectral images of 10 T. galloi specimens mixed with two T. atopivirilia specimens, two T. atopivirilia specimens were correctly identified in all validation samples, but 36% of the adult specimens were incorrectly classified as T. pretiosum (although they were T. galloi specimens). Thus, the classification model accurately detected whether T. atopivirilia individuals were present or absent in the mixtures (classification accuracy=100%). However, the classification model was not able to accurately differentiate between adult specimens of T. galloi and T. pretiosum (classification accuracy <65.0%).

Table 1.

Linear discriminant classification of parasitized host eggs

Linear discriminant classification of parasitized host eggs
Linear discriminant classification of parasitized host eggs
Fig. 3.

Adult Trichogramma individuals. (A) Images of adult Trichogramma specimens. (B) Relative reflectance from host eggs in 160 spectral bands from 405 to 905 nm.

Fig. 3.

Adult Trichogramma individuals. (A) Images of adult Trichogramma specimens. (B) Relative reflectance from host eggs in 160 spectral bands from 405 to 905 nm.

DISCUSSION

We are unaware of any published studies on the cuticular hydrocarbon composition of Trichogramma species. However, several studies have described how cuticular hydrocarbon composition of adult individuals of other minute parasitoids is influenced by age, mating status and diet, and how it also may vary among males and females (Howard and Baker, 2003; Howard and Pérez-Lachaud, 2002). Thus, our reflectance-based profiling of parasitoid species is likely associated with species-specific variation in cuticular composition. Compared with classical taxonomy (under the microscope) or molecular-based classification of minute and closely related animals, plant seeds and growing plants, a reflectance-based method may be of considerable relevance to a wide range of biological studies. There are numerous approaches to classification of hyperspectral imaging data, and only recently has the approach (a combination of variogram analysis and linear discriminant analysis) used in this study been described and successfully applied (Nansen, 2012; Nansen et al., 2010a; Nansen et al., 2009; Nansen et al., 2014). However, this analysis represents the first application of this classification method to the identification of animals. There appear to be two main advantages of the proposed method. (1) The classification model only requires reflectance values in a few narrow spectral bands. This means less computer-processing requirements than when support vector machine (Vapnik, 1995) or other computer-intensive classification methods are deployed. (2) Incorporation of information about the spatial data structure (variogram parameters derived from variogram analysis) appears to provide substantial accuracy and robustness to the classification model. The latter point implies that the proposed classification method relies on reflectance data being acquired at a fairly high spatial resolution. The current study was based on data acquired at a spatial resolution of 45 pixels mm−2. We found clear spectral trends, with reflectance data acquired from certain spectral ranges providing better separation than in other spectral ranges. We also demonstrated that classification of parasitized host eggs and adult Trichogramma specimens required reflectance data in different spectral bands. Parasitized host eggs were classified with high accuracy (>90% classification accuracy). Regarding adult specimens, we were able to accurately separate adult T. atopivirilia specimens from T. galloi and T. pretiosum, but the latter two species could not be accurately classified. Interestingly, Almeida (Almeida, 2004) developed a dichotomous key of 17 South American Trichogramma species based on the DNA sequence of the internally transcribed spacer ITS2 and (Fig. 4). Almeida (Almeida, 2004) found that the ITS2 product size varied from 379 to 632 bp among species, and its size successfully grouped the species into two groups (this was the first and therefore most important identification trait). This key suggested that T. galloi and T. pretiosum are closely related and comparatively distant from T. atopovirilia, as the latter species is separated from the other two species at the first dichotomous level (whether the size of the PCR product is above or below 620 bp). This may indicate that the difficulty associated with the separation of T. galloi and T. pretiosum may be linked to their phylogenetic proximity. Our results emphasize the importance of using high-spectral and high-spatial resolution data in the classification of closely related organisms, and hyperspectral imaging may be of relevance to a wide range of commercial (i.e. producers of biocontrol agents), taxonomic and evolutionary research applications.

Table 2.

Linear discriminant classification of adult Trichogramma specimens

Linear discriminant classification of adult Trichogramma specimens
Linear discriminant classification of adult Trichogramma specimens

MATERIALS AND METHODS

Insects

Newly oviposited Mediterranean flour moth eggs (0–24 h) are here referred to as host eggs, and they were glued (Arabic gum, Goyana) individually onto a piece of pink paper (4×0.5 cm), referred to as egg cards (Fig. 1A). Gluing host eggs to cards does not affect developing Trichogramma parasitoids, and most commercial shipments of Trichogramma parasitoids consist of parasitized host eggs glued onto egg cards. Each egg card was exposed for 12 h to parasitism by a single Trichogramma female inside a 1.5 cm (diameter) by 7 cm (length) glass vial maintained at 25±1°C, relative humidity 70±10% and 14 h:10 h light:dark light program. One droplet of honey was placed inside the vial as a food source for each Trichogramma female. As a control, we also prepared egg cards without parasitism. Hyperspectral images were acquired from egg cards at 24 h intervals after parasitism during six consecutive days. Hyperspectral images acquired from non-parasitized eggs were only used to develop a spectral filter to exclude non-parasitized eggs from the classification (see Eqn 1 below). We also acquired hyperspectral images of samples of male and female adult parasitoids of all three Trichogramma species, which had been killed in 80% ethanol, dried and placed on a white piece of paper. Trichogramma species identification was based on Querino and Zucchi (Querino and Zucchi, 2011).

Fig. 4.

Phylogenetic tree of Trichogramma species. The tree (Almeida, 2004) of 17 South American Trichogramma species is based on the size (bp) and three endonucleases (EcoRI, MseI and MaeI) of the DNA sequence entitled internally transcribed spacer ITS2 (Stouthamer et al., 1999). The three species included in this study are highlighted (T. atopovirilia, T. galloi and T. pretiosum).

Fig. 4.

Phylogenetic tree of Trichogramma species. The tree (Almeida, 2004) of 17 South American Trichogramma species is based on the size (bp) and three endonucleases (EcoRI, MseI and MaeI) of the DNA sequence entitled internally transcribed spacer ITS2 (Stouthamer et al., 1999). The three species included in this study are highlighted (T. atopovirilia, T. galloi and T. pretiosum).

Hyperspectral imaging

Similar to previously published studies (Nansen et al., 2010a; Nansen et al., 2010b; Nansen et al., 2008), we used a hyperspectral push broom spectral camera (PIKA II, Resonon Inc., Bozeman, MT, USA), which collects 160 bands in the range from 405 to 907 nm. The objective lens had a 35 mm focal length (maximum aperture of F1.4) and was optimized for the visible and NIR spectra. The main specifications of the spectral camera are as follows: interface, Firewire (IEEE 1394b); output, digital (12 bit); 160 bands (spectral) by 640 pixels (spatial); angular field of view, 7 deg; and spectral resolution, <3 nm. All hyperspectral images were collected with artificial lighting from 15 W, 12 V LED light bulbs mounted in two angled rows, one on either side of the lens, with three bulbs in each row. A voltage stabilizer (Tripp-Lite, PR-7b, www.radioreference.com) powered the lighting. Ambient climate conditions were between 19 and 22°C and between 30% and 40% relative humidity. A piece of white Teflon (K-Mac Plastics, MI, USA) was used for white calibration, and ‘relative reflectance’ refers to proportional reflectance compared with that obtained from Teflon. Consequently, relative reflectance values ranged from 0 to 1. Hyperspectral images of host eggs and adult Trichogramma specimens were acquired at a spatial resolution of 45 pixels mm−2.

Reflectance data processing and analysis

A customized software package was used to convert hyperspectral image files in BIL-format into txt-files, and these were subsequently imported into PC-SAS 9.2 (SAS Institute, NC, USA) for processing and analysis. As part of omitting the pink background, non-parasitized eggs and stochastic noise from all hyperspectral images of host eggs (Fig. 1A), we applied a radiometric filter (Eqn 1), so that reflectance profiles were only included if the relative reflectance in the spectral band at 700 nm (R700) was:
formula
(1)
Fig. 5.

Variogram analysis. The fit describes the spatial structure (variogram analysis) of reflectance data with: (1) the a parameter denoting the intercept, (2) the b parameter denoting the asymptote and (3) the c parameter denoting the slope of the increase towards the asymptote.

Fig. 5.

Variogram analysis. The fit describes the spatial structure (variogram analysis) of reflectance data with: (1) the a parameter denoting the intercept, (2) the b parameter denoting the asymptote and (3) the c parameter denoting the slope of the increase towards the asymptote.

A different radiometric filter was applied to hyperspectral images of Trichogramma adults (Fig. 3A, Eqn 2), so that reflectance profiles were only included if the relative reflectance in the spectral band at 544 m (R544) was:
formula
(2)

With a spatial resolution of 45 pixels mm−2 and after radiometric filtering, we obtained, on average, 108.13±4.45 pixels per host egg and 52.65±1.62 pixels per adult Trichogramma specimen. Analysis of variance (PROC MIXED) was used to compare averages of pixels acquired from adult male and female specimens.

The classification of reflectance data was based on a combination of variogram analysis (Nansen, 2012; Nansen et al., 2014) and linear discriminant analysis (Fisher, 1936). In brief, the analytical approach consists of conducting variogram analysis of reflectance values in individual spectral bands from each hyperspectral image. We conducted variogram analysis (PROC VARIOGRAM) using the following settings for host eggs: lagdistance=1, maxlags=10 and outpdistance=10, and lagdistance=1, maxlags=5 and outpdistance=5 for adult Trichogramma specimens. The difference in the lagdistance number was based on host eggs being considerably larger than adult specimens. Then, a non-linear regression (Eqn 3) is fitted to each of the output variogram data:
formula
(3)
in which a, b and c are fitted parameters, and F(D) is the semi-variance at each lag distance interval, D. As illustrated in Fig. 5: (1) the a parameter denotes the intercept, (2) the b parameter denotes the asymptote and (3) the c parameter denotes the slope of the increase towards the asymptote. Thus, the proposed classification approach is based on the assumption that the three variogram parameters (a, b and c) can be used to describe the spatial structure of the reflectance data in the single reflectance band. The model training data sets consisted of 60 parasitized host eggs (20 of each of the three Trichogramma species) and 120 adult Trichogramma specimens (20 of each combination of Trichogramma species and sex). Variogram analysis was conducted in individual spectral bands in 10-band intervals (16 spectral bands), so a total of 2880 variogram analyses were conducted to develop the classification models of parasitized host eggs and for adult specimens [(60 hyperspectral images of host eggs + 120 hyperspectral images of adult specimens)×16 spectral bands]. Subsequently, the derived variogram parameters (a, b and c) and average reflectance values in the same spectral band were used as input variables in a linear stepwise discriminant analysis (PROC STEPWISE) of host eggs or adult specimens selecting the combination of variogram parameters and spectral band with the best separation of classification classes (species of parasitoids). For each of the two classifications (host eggs and adult specimens), linear discriminant analysis (PROC DISCRIM option=crossvalidate) with jack-knife cross-validation was used to assess the classification accuracy of each of the two data sets. Jack-knife cross-validation implies removal of one sample and its use for independent validation, repeating this procedure with all samples. As a second validation approach, the classification models were used to classify independent hyperspectral imaging data from 25 parasitized host eggs (12 host eggs parasitized by T. atopovirilia and 13 host eggs parasitized by T. pretiosum) and from nine mixtures of adult Trichogramma specimens. These mixtures of adult Trichogramma specimens consisted of 10 T. galloi only or 10 T. galloi individuals mixed with two specimens of T. atopovirilia or T. pretiosum, and there were three replicated hyperspectral images for each combination.

FOOTNOTES

Funding

We thank the National Council of Technological and Scientific Development (CNPq) and São Paulo research foundation (Fapesp) for part of financial support.

References

Almeida
R. P.
(
2004
).
Trichogramma and its relationship with Wolbachia: identification of Trichogramma species, phylogeny, transfer and costs of Wolbachia symbionts
.
PhD thesis
,
Wageningen University
,
The Netherlands
.
Anderson
K.
,
Gaston
K. J.
(
2013
).
Lightweight unmanned aerial vehicles will revolutionize spatial ecology
.
Front. Ecol. Environ
11
,
138
-
146
.
Bigler
F.
(
1989
).
Quality assessment and control in entomophagous insects used for biological control
.
J. Appl. Entomol.
108
,
390
-
400
.
Butler
S. M.
,
Moon
R. D.
,
Hinkle
N. C.
,
Millar
J. G.
,
McElfresh
J. S.
,
Mullens
B. A.
(
2009
).
Characterization of age and cuticular hydrocarbon variation in mating pairs of house fly, Musca domestica, collected in the field
.
Med. Vet. Entomol.
23
,
426
-
442
.
Cerutti
F.
,
Bigler
F.
(
1995
).
Quality assessment of Trichogramma brassicae in the laboratory
.
Entomol. Exp. Appl.
75
,
19
-
26
.
Chiao
C. C.
,
Wu
W.-Y.
,
Chen
S.-H.
,
Yang
E.-C.
(
2009
).
Visualization of the spatial and spectral signals of orb-weaving spiders, Nephila pilipes, through the eyes of a honeybee
.
J. Exp. Biol.
212
,
2269
-
2278
.
Chiao
C. C.
,
Wickiser
J. K.
,
Allen
J. J.
,
Genter
B.
,
Hanlon
R. T.
(
2011
).
Hyperspectral imaging of cuttlefish camouflage indicates good color match in the eyes of fish predators
.
Proc. Natl. Acad. Sci. USA
108
,
9148
-
9153
.
Clark
R. N.
(
1999
).
Spectroscopy of rocks and minerals, and principles of spectroscopy
. In
Manual of Remote Sensing
, Vol.
3
(ed.
Rencz
A. N.
), pp.
3
-
58
.
New York, NY
:
John Wiley and Sons
.
Danks
H. V.
(
1988
).
Systematics in support of entomology
.
Annu. Rev. Entomol.
33
,
271
-
294
.
De Loof
A.
,
Huybrechts
J.
,
Geens
M.
,
Vandersmissen
T.
,
Boerjan
B.
,
Schoofs
L.
(
2010
).
Sexual differentiation in adult insects: male-specific cuticular yellowing in Schistocerca gregaria as a model for reevaluating some current (neuro)endocrine concepts
.
J. Insect Physiol.
56
,
919
-
925
.
Dowell
F. E.
,
Throne
J. E.
,
Wang
D.
,
Baker
J. E.
(
1999
).
Identifying stored-grain insects using near-infrared spectroscopy
.
J. Econ. Entomol.
92
,
165
-
169
.
Ferreira-Caliman
M. J.
,
Nascimento
F. S.
,
Turatti
I. C.
,
Mateus
S.
,
Lopes
N. P.
,
Zucchi
R.
(
2010
).
The cuticular hydrocarbons profiles in the stingless bee Melipona marginata reflect task-related differences
.
J. Insect Physiol.
56
,
800
-
804
.
Fisher
R. A.
(
1936
).
The use of multiple measurements in taxonomic problems
.
Ann. Eugen.
7
,
179
-
188
.
Geiselhardt
S.
,
Otte
T.
,
Hilker
M.
(
2009
).
The role of cuticular hydrocarbons in male mating behavior of the mustard leaf beetle, Phaedon cochleariae (F.)
.
J. Chem. Ecol.
35
,
1162
-
1171
.
Hassan
S. A.
(
1997
).
Criação da traça do milho, Sitotroga cerealella, para a produção massal de Trichogramma
. In
Trichogramma e o Controle Biológico Aplicado
(ed.
Parra
J. R. P.
,
Zucchi
R. A.
), pp.
173
-
182
.
Piracicaba, Sao Paulo, Brazil
:
FEALQ
.
Howard
R. W.
,
Baker
J. E.
(
2003
).
Cuticular hydrocarbons and wax esters of the ectoparasitoid Habrobracon hebetor: ontogenetic, reproductive, and nutritional effects
.
Arch. Insect Biochem. Physiol.
53
,
1
-
18
.
Howard
R. W.
,
Pérez-Lachaud
G.
(
2002
).
Cuticular hydrocarbons of the ectoparasitic wasp Cephalonomia hyalinipennis (Hymenoptera: Bethylidae) and its alternative host, the stored product pest Caulophilus oryzae (Coleoptera: Curculionidae)
.
Arch. Insect Biochem. Physiol.
50
,
75
-
84
.
Jia
F.
,
Maghirang
E.
,
Dowell
F.
,
Abel
C.
,
Ramaswamy
S.
(
2007
).
Differentiating tobacco budworm and corn earworm using near-infrared spectroscopy
.
J. Econ. Entomol.
100
,
759
-
764
.
Klarica
J.
,
Bittner
L.
,
Pallua
J.
,
Pezzei
C.
,
Huck-Pezzei
V.
,
Dowell
F.
,
Schied
J.
,
Bonn
G. K.
,
Huck
C.
,
Schlick-Steiner
B. C.
, et al. 
. (
2011
).
Near-infrared imaging spectroscopy as a tool to discriminate two cryptic Tetramorium ant species
.
J. Chem. Ecol.
37
,
549
-
552
.
Kuhlmann
U.
,
Mills
N. J.
(
1999
).
Comparative analysis of the reproductive attributes of three commercially produced Trichogramma species (Hymenoptera: Trichogrammatidae)
.
Biocontrol Sci. Technol.
9
,
335
-
346
.
Lapointe
S. L.
,
Hunter
W. B.
,
Alessandro
R. T.
(
2004
).
Cuticular hydrocarbons on elytra of the Diaprepes root weevil Diaprepes abbreviatus (L.) (Coleoptera: Curculionidae)
.
Agric. For. Entomol.
6
,
251
-
257
.
Lenteren
J. C. v.
(
2008
).
Current Situation of Biological Control
.
Wageningen, The Netherlands
:
IOBC
.
Lewis
W. J.
(
1976
).
Production and performance of Trichogramma reared on eggs of Heliothis zea and other hosts
.
Environ. Entomol.
5
,
449
-
452
.
Li
L. Y.
(
1994
).
Worldwide use of Trichogramma for biological control on different crops: a survey
. In
Biological Control with Egg Parasitoids
(ed.
Wajnberg
E.
,
Hassan
S. A.
), pp.
37
-
53
.
Wallingford, UK
:
CAB International
.
Losey
J. E.
,
Calvin
D. D.
(
1995
).
Quality assessment of four commercially available species of Trichogramma (Hymenoptera: Trichogrammatidae)
.
J. Econ. Entomol.
88
,
1243
-
1250
.
Lundgren
J. G.
,
Heimpel
G. E.
(
2003
).
Quality assessment of three species of commercially produced Trichogramma and the first report of thelytoky in commercially produced Trichogramma
.
Biol. Control
26
,
68
-
73
.
Nansen
C.
(
2012
).
Use of variogram parameters in analysis of hyperspectral imaging data acquired from dual-stressed crop leaves
.
Remote Sensing
4
,
180
-
193
.
Nansen
C.
,
Kolomiets
M.
,
Gao
X.
(
2008
).
Considerations regarding the use of hyperspectral imaging data in classifications of food products, exemplified by analysis of maize kernels
.
J. Agric. Food Chem.
56
,
2933
-
2938
.
Nansen
C.
,
Macedo
T.
,
Swanson
R.
,
Weaver
D. K.
(
2009
).
Use of spatial structure analysis of hyperspectral data cubes for detection of insect-induced stress in wheat plants
.
Int. J. Remote Sens.
30
,
2447
-
2464
.
Nansen
C.
,
Abidi
N.
,
Sidumo
A. J.
,
Gharalari
A. H.
(
2010a
).
Using spatial structure analysis of hyperspectral imaging data and fourier transformed infrared analysis to determine bioactivity of surface pesticide treatment
.
Remote Sens.
2
,
908
-
925
.
Nansen
C.
,
Herrman
T.
,
Swanson
R.
(
2010b
).
Machine vision detection of bonemeal in animal feed samples
.
Appl. Spectrosc.
64
,
637
-
643
.
Nansen
C.
,
Zhang
X.
,
Aryamanesh
N.
,
Yan
G.
(
2014
).
Use of variogram analysis to classify field peas with and without internal defects caused by weevil infestation
.
J. Food Eng.
123
,
17
-
22
.
Nunes
T. M.
,
Turatti
I. C. C.
,
Mateus
S.
,
Nascimento
F. S.
,
Lopes
N. P.
,
Zucchi
R.
(
2009
).
Cuticular hydrocarbons in the stingless bee Schwarziana quadripunctata (Hymenoptera, Apidae, Meliponini): differences between colonies, castes and age
.
Genet. Mol. Res.
8
,
589
-
595
.
Parra
J. R. P.
(
2010
).
Mass rearing of egg parasitoids for biological control programs
In
Egg Parasitoids in Agroecosystems with Emphasis on Trichogramma
(ed.
Consoli
F. L.
,
Parra
J. R. P.
,
Zucchi
R. A.
), pp.
267
-
278
.
Dordrecht, Germany
:
Springer
.
Pinto
J. D.
(
2006
).
A review of the New World genera of Trichogrammatidae (Hymenoptera)
.
J. Hymenoptera Res.
15
,
38
-
163
.
Pinto
J. D.
,
Stouthamer
R.
(
1994
).
Systematics of the Trichogrammatidae with emphasis on Trichogramma
. In
Biological Control with Egg Parasitoids
(ed.
Wajnberg
E.
,
Hassan
S. A.
), pp.
1
-
36
.
Wallingford, UK
:
Cab International
.
Querino
R. B.
,
Zucchi
R. A.
(
2011
).
Guia de Identificação de Trichogramma Para o Brasil
.
Brasilia, Brazil
:
Embrapa
.
Reeves
W. K.
,
Peiris
K. H. S.
,
Scholte
E.-J.
,
Wirtz
R. A.
,
Dowell
F. E.
(
2010
).
Age-grading the biting midge Culicoides sonorensis using near-infrared spectroscopy
.
Med. Vet. Entomol.
24
,
32
-
37
.
Roux
O.
,
Gers
C.
,
Legal
L.
(
2008
).
Ontogenetic study of three Calliphoridae of forensic importance through cuticular hydrocarbon analysis
.
Med. Vet. Entomol.
22
,
309
-
317
.
Singh
C. B.
,
Jayas
D. S.
,
Paliwal
J.
,
White
N. D. G.
(
2010
).
Identification of insect-damaged wheat kernels using short-wave near-infrared hyperspectral and digital colour imaging
.
Computers and Electronics in Agriculture
73
,
118
-
125
.
Smith
S. M.
(
1996
).
Biological control with Trichogramma: advances, successes, and potential of their use
.
Annu. Rev. Entomol.
41
,
375
-
406
.
Steiner
S.
,
Mumm
R.
,
Ruther
J.
(
2007
).
Courtship pheromones in parasitic wasps: comparison of bioactive and inactive hydrocarbon profiles by multivariate statistical methods
.
J. Chem. Ecol.
33
,
825
-
838
.
Stouthamer
R.
,
Hu
J.
,
van Kam
F. J. P. M.
,
Platner
G. R.
,
Pinto
J. D.
(
1999
).
The utility of internally transcribed spacer 2 DNA sequences of the nuclear ribosomal gene for distinguishing sibling species of Trichogramma
.
BioControl
43
,
421
-
440
.
Sweeney
A. M.
,
Boch
C. A.
,
Johnsen
S.
,
Morse
D. E.
(
2011
).
Twilight spectral dynamics and the coral reef invertebrate spawning response
.
J. Exp. Biol.
214
,
770
-
777
.
Vapnik
V.
(
1995
).
The Nature of Statistical Learning Theory
.
New York, NY
:
Springer
.
Zeil
J.
,
Hofmann
M.
(
2001
).
Signals from ‘crabworld’: cuticular reflections in a fiddler crab colony
.
J. Exp. Biol.
204
,
2561
-
2569
.
Zhu
G. H.
,
Ye
G. Y.
,
Hu
C.
,
Xu
X. H.
,
Li
K.
(
2006
).
Development changes of cuticular hydrocarbons in Chrysomya rufifacies larvae: potential for determining larval age
.
Med. Vet. Entomol.
20
,
438
-
444
.
Zucchi
R. A.
,
Querino
R. B.
(
2010
).
Diversity and hosts of Trichogramma in the New World, with emphasis in South America
In
Egg Parasitoids in Agroecosystems with Emphasis on Trichogramma
(ed.
Consoli
F. L.
,
Parra
J. R. P.
,
Zucchi
R.
), pp.
219
-
236
.
The Netherlands
:
Springer
.