Landscape Effects on Reproduction of Euschistus servus (Hemiptera: Pentatomidae), a Mobile, Polyphagous, Multivoltine Arthropod Herbivore

Landscape Effects on Reproduction of Euschistus servus (Hemiptera: Pentatomidae), a Mobile,... Abstract Landscape factors can significantly influence arthropod populations. The economically important brown stink bug, Euschistus servus (Say) (Hemiptera: Pentatomidae), is a native mobile, polyphagous and multivoltine pest of many crops in southeastern United States and understanding the relative influence of local and landscape factors on their reproduction may facilitate population management. Finite rate of population increase (λ) was estimated in four major crop hosts—maize, peanut, cotton, and soybean—over 3 yr in 16 landscapes of southern Georgia. A geographic information system (GIS) was used to characterize the surrounding landscape structure. LASSO regression was used to identify the subset of local and landscape characteristics and predator densities that account for variation in λ. The percentage area of maize, peanut and woodland and pasture in the landscape and the connectivity of cropland had no influence on E. servus λ. The best model for explaining variation in λ included only four predictor variables: whether or not the sampled field was a soybean field, mean natural enemy density in the field, percentage area of cotton in the landscape and the percentage area of soybean in the landscape. Soybean was the single most important variable for determining E. servus λ, with much greater reproduction in soybean fields than in other crop species. Penalized regression and post-selection inference provide conservative estimates of the landscape-scale determinants of E. servus reproduction and indicate that a relatively simple set of in-field and landscape variables influences reproduction in this species. brown stink bug, least angle regression There is increasing evidence that community structure, species abundance, and biotic interactions of invertebrate species in farmlands are influenced by larger-scale processes occurring at the landscape level (habitat size, spatial arrangement, connectivity and quality, and landscape matrix: Andow 1983, Marino and Landis 1996, Colunga-Garcia et al. 1997, Thies et al. 2003, Tscharntke and Brandl 2004, Schweiger et al. 2005, Bianchi et al. 2006, Tscharntke et al. 2007, Gardiner et al. 2009, Yasuda et al. 2011). Arable landscapes are often intensely managed and frequent application of agrochemicals can be a significant cause of biodiversity loss (e.g., Matson et al. 1997, Wilson et al. 1999). Further, structurally more complex landscapes—i.e., those with higher amounts of non-crop area such as woodland, hedgerows, grassland, fallows, and pastures—may compensate for locally reduced diversity inside intensively managed crop fields mainly through rapid re-colonization of resource-rich crop fields by highly dispersive organisms (Rusch et al. 2016). These non-crop areas can also provide insects with overwintering sites, summer aestivation sites, resting sites, mating sites, and sites that have spatially separated resources that are required to meet their needs (Holland and Fahrig 2000, Tscharntke et al. 2012). Additionally, spillover across habitats often increases with increasing edge density, or perimeter-area ratios, which can enhance or inhibit functional connectivity among habitats (Olson and Andow 2008, Tscharntke et al. 2012). Further, Sivakoff et al. (2013) and Meisner et al. (2017) found that crop composition immediately adjacent to a cotton field was associated with substantial differences in cotton yield, the pest species Lygus hesperus Knight (Hemiptera: Miridae) density and pesticide use, suggesting that spillover effects of arthropod species among crops of different quality may also occur in landscapes. The ability of a population to persist and increase after colonizing habitats is reflected in its finite rate of population increase (λ) in that habitat. Factors that influence λ include micro- and macro-climates, resource availability, competition, predation, and dispersal (Norton et al. 2005). Knowledge of the local and landscape variables influencing λ in different landscapes may lead to a better understanding of the factor(s) contributing to population buildup in specific areas. The southern green stink bug Nezara viridula (L.) (Hemiptera: Pentatomidae), the brown stink bug Euschistus servus (Say) (Hemiptera: Pentatomidae), and the green stink bug Chinavia hilaris (Say) (Hemiptera: Pentatomidae) are important agricultural pests in southeastern United States, that became more prominent in cotton after widespread adoption of Bt cotton and eradication of the cotton boll weevil (Turnipseed et al. 1995, Greene et al. 2006, Zeilinger et al. 2011). The major row crops economically damaged by stink bugs are field and sweet corn, soybean and cotton (McPherson and McPherson 2000; Koch and Pahs 2014, 2015; Soria et al. 2017). Although peanut—another major crop in the Southeast—is also a host of stink bugs (Tillman et al. 2009), they have not been reported to cause economic damage to this crop. In this study we concentrate on E. servus because during the last several years it has become the dominant stink bug species in southern Georgia (Olson et al. 2012). E. servus is a native species occurring from the southeastern United States west through Louisiana, Texas, New Mexico, and Arizona into California (McPherson 1982). It is bivoltine throughout its range, and overwinters in the adult stage under crop residue, leaves, pieces of bark, and in bunches of grass, preferring open fields (McPherson and McPherson 2000). This species is highly mobile and polyphagous and prefers feeding on the seeds/fruit of host plants (McPherson and McPherson 2000); thus, they move among host plants in response to changing phenology of the hosts (McPherson and McPherson 2000, Blinka 2008). Maize is an early planted host with some overlap in occurrence with later-planted peanut, cotton and soybean. All of these crops are reproductive hosts for E. servus (Herbert and Toews 2011, Koch and Pahs 2014), colonization preference for soybean is much higher than for peanut and cotton (Olson et al. 2011) and peanut is a poorer quality host in terms of adult longevity than are cotton and soybean (Olson et al. 2016). In addition, numerous non-crop E. servus hosts can exist in non-woodland and woodland field borders surrounding the crops (McPherson and McPherson 2000), and are suspected to be sources of stink bugs to the adjacent crops in the spring (Reay-Jones 2010, Olson et al. 2012, Tillman et al. 2014). Evidence for the importance of landscape-level determinants of natural enemy populations and their role in natural pest control is increasing (Werling et al. 2011, Avelino 2012, Fabian et al. 2013, Rusch et al. 2016). The role of natural enemies in stink bug population dynamics remains poorly understood, at either the field or landscape level. Olson and Ruberson (2012) indicated the importance of fire ants (Solenopsis invicta Buren, Hymenoptera: Formicidae) as predators of stink bug eggs in unsprayed cotton and peanut, whereas long-horned grasshoppers (Orthoptera: Tettigoniidae) were dominant egg predators in unsprayed soybean. Geocoris spp. (Hemiptera: Geocoridae) also feed on stink bug eggs (Olson and Ruberson 2012) and are abundant predators in cotton and soybean (Naranjo and Simac 1985, Pfannenstiel and Yeargan 1998). There is little known of the ecology of these predators in agroecosystems with respect to growth of E. servus populations. Given the variation among crop species for E. servus host quality and predation rates, a landscape approach to the study of E. servus populations encompassing major crop and non-crop hosts may reveal population patterns that can account for the present-day outbreaks in cotton and other crops. The cross-habitat spillover hypothesis put forth by Tshcarntke et al. (2012) suggests that more mobile species and species that need multiple cover types may spillover and flourish in landscapes with high functional connectivity. Therefore, we tested the hypotheses 1a) that as the percentage area of woodland and pasture, maize, peanut, cotton and/or soybean increases in the landscape, the net reproduction of E. servus increases in the landscape, 1b) as the number of maize, peanut, cotton and soybean fields closest to the focal fields increases, the net reproduction of E. servus increases in the landscape, and 2) higher densities of predators—specifically, fire ants, longhorned grasshoppers and Geocoris spp. in sampled fields reduces the net reproduction of E. servus in that field. Materials and Methods Sampling Plan We identified two areas (Southwest and East-Central) of the Coastal Plain of Georgia (Fig. 1) which to sample stink bugs in crop landscapes during the years 2009–2011. The areas were separated by approximately 150 km. Within each area, we randomly identified two or three 4.8 × 4.8 km (2,330 ha) landscapes, with two landscapes in 2011 in the Southwest area being 5.3 × 8.3 km (4,399 ha) to encompass enough of the required fields. The Southwest area had two landscape samples during 2009 (designated Shirah and Baggs), and three landscape samples during 2010 and 2011 (Vinwell, Wright and Moultrie). The East-Central area had two landscapes sampled during 2009 (North and South), and three landscape samples during 2010 (Davis, Henderson and Rufus) and 2011 (Henderson, Irwin and Rufus). Landscape samples differed in location each year because of shifting crop patterns. Working with local landowners in each landscape, we identified three fields of maize, peanut, soybean, and cotton. Each crop field was sampled weekly, from early June to August in maize, and from mid-July to late September for the other crops. The fields were commercial fields that were managed in accordance with the growers’ practices. We removed 43 fields where E. servus λ could not be estimated because of low numbers and outlier soybean fields that had a high frequency of insecticide applications by two growers in eight soybean fields (4–6 applications) or crop failure (plants less than 30 cm in height) in one grower’s three small fields (≤0.40 ha). This resulted in a total of 138 fields used in the analyses over the 3-yr period of this study. Fig. 1. View largeDownload slide A 2009 landscape within the Coastal Plain of Georgia illustrating crop types and sampled sites. Fig. 1. View largeDownload slide A 2009 landscape within the Coastal Plain of Georgia illustrating crop types and sampled sites. Each field was sampled using two permanent parallel transects (spaced 30.5 m apart) running perpendicular to the edge of the field. All sampled field edges were adjacent to woodland. A total of 20 sampling points in 2009 and 15 sampling points in 2010 and 2011 were established along each transect. The first sample point was placed 1 m from the crop edge for all years and in 2009, the next 19 samples were spaced at 5 m intervals, whereas in 2010 and 2011, samples 2 through 10 were spaced at 5 m intervals and the last five samples were spaced at 10 m intervals (=101 m from crop edge for all fields). Samples alternated weekly from crop rows 1–5 to the left and the right sides of transects to reduce repetitive plant and population disturbance. Sampling of maize was done using a two-person, whole-plant visual count with the samplers on opposite sides of the maize row for a total sampling distance of 1.5 m (ca 8 plants per sample). Peanut was sampled using a Vortis suction sampler (15 cm diameter inlet, Burkard Manufacturing Company, Ltd., Hertfordshire, UK) with 12 × 3-s suctions at 7–8 cm from the soil at each sampling point (2,121 cm2 sampled per sampling station). Cotton and soybean were sampled using a 1.5 m drop cloth with 10 shakes per sample. In all crops, about 1.5 linear row-meters of plants were sampled at each point. All stink bugs were identified and assigned to a developmental stage, but only E. servus was common enough over the 3 yr to analyze in detail. Landscape Characteristics For each landscape, we used a geographic information system (GIS) to determine the percentage of area in the landscape that contained E. servus host crops (intensively produced maize, peanut, cotton and soybean); semi-natural habitat consisting of woodland and pasture or ‘green-veining’ (GV), and the number of maize, peanut, cotton and soybean fields within 100, 500, and 1,000 m radii from the focal sampling field as a measure of cropland connectivity. Assuming no physical barriers to E. servus movement, we calculated straight-line connections from the edges of sampled crop fields to the closest edges of nearby crops. Females of an ecologically similar species, N. viridula, have been observed dispersing up to 1,000 m per day by flight in search of feeding or oviposition sites (Kiritani and Sasaba 1969). Natural Enemies We included S. invicta, adults and nymphs of longhorned grasshoppers, and adults and nymphs of Geocoris spp. in our weekly samples. In addition to generalist predator sampling, we estimated parasitism rates by collecting and incubating a subsample of stink bug adults and nymphs. Numbers of natural enemies and parasitism rates were averaged over field samples to coincide with our estimates of λ per field. However, because parasitism of adults (1.4%) and nymphs (0%) was very low, overall parasitism was excluded from subsequent analysis. Longhorned grasshopper density was very low so they were also excluded from subsequent analysis. Relative Finite Rate of Population Increase Relative finite rate of population increase (λ generation−1) was estimated for each sampled field where adults were observed in the field. The date that adults were first observed was considered to be the date of first colonization and the dates that peaks in adult density occurred were noted. If late-instar nymphs were observed before adults were observed, the date of first adult colonization was assumed to have occurred 20 d before the date of first sighting of nymphs, based on E. servus developmental time (Munyaneza & McPherson 1994). On average over all crops, the number of days from the first adult peak to the second adult peak was 39.7 ± 2.5 d, so during the 40 d following the date of first adult colonization, any adults observed were considered to be colonists. After this period, adults were considered to be progeny of the colonists. Eggs were too infrequently sampled to be used to estimate an oviposition period or an initial density of the offspring generation, and nymphs were found less frequently than adults. Thus, we used colonizing and progeny adults to estimate λ. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). An example is shown in Fig. 2. Fig. 2. View largeDownload slide An example of incidence data: the average number of brown stink bug adults, nymphs, and eggs per sample versus day of year in maize in the North Coffee landscape of the East-Central region. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). Fig. 2. View largeDownload slide An example of incidence data: the average number of brown stink bug adults, nymphs, and eggs per sample versus day of year in maize in the North Coffee landscape of the East-Central region. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). Statistical Analyses First, we investigated cropland and non-cropland heterogeneity over years. We analyzed the effects of year on arcsine-square root transformed percentage area of cropland hosts, GV (woodland and pasture), as well as the perimeter-to-area ratio of the sampled fields, and the number of cropland fields within 100, 500, and 1,000 m from the sampled field with ANOVA and used Tukey’s HSD to separate the means. Second, we investigated which factors explained E. servus population increase, λ. We included three factors in this analysis: 3 yr, four or six landscapes (landscapes varied over years), and three maize, peanut, cotton and soybean fields in each landscape. During 2009, four landscapes were sampled and during 2010 and 2011 six landscapes were sampled. The design was a nested ANOVA with years as whole plot factors and whole plot error defined by landscapes within years (error 1). The subplots were the crops (and crop interactions with year), and groups were the unit of replication (error 2) after removing all landscape interactions with crops (and landscape by crop interactions with year). The natural logarithm of λ (lnλ) was analyzed as the untransformed λ was highly skewed and heteroscedastic. The standardized residuals for error 2 were normal (Supplementary Fig. S1a and b), and there was no correlation between the standardized residuals and the predicted values (Supplementary Fig. S2) thereby, satisfying assumptions of an ANOVA. However, lnλ error was still slightly heteroscedastic, so we also conducted a randomization test, using the same ANOVA model. Values of lnλ were randomized among crops and groups within landscapes, and landscapes were randomized among years, making sure that the unbalanced structure was preserved. These randomizations independently permuted the units of replication with respect to year at the whole plot level and with respect to crop (and its interactions) at the subplot level. The estimated F-value from each randomization for each term of interest was used to construct null distributions, and the observed F was used to calculate the P-value. The simulations were repeated 2.5 million times so that the estimated P-value was accurate to five significant digits. Third, we determined the set of landscape and local scale variables that best explained variation in E. servus reproduction using the LASSO method of variable selection (Hastie, Tibshirani, and Friedman 2009). Stepwise regression and model averaging are the other two ways to select a subset of variables. Stepwise regression generally overestimates parameter values, and underestimates standard errors. Model averaging requires a priori specification of a set of models. The LASSO procedure uses a regularization penalty to ‘shrink’ coefficient estimates of a linear combination of covariates, using a pre-determined estimate of the regulator or penalty parameter, γ1 (note, while other authors have represented the LASSO regulator as λ, we adopt the notation of Hooten and Hobbs (2015) to differentiate it from our estimates of stink bug reproduction, λ). The LASSO procedure is a special case of the elastic net method (Hastie et al. 2009); we conducted K-fold cross-validation of the elastic net tuning parameter, α (results not shown) the minimum mean-squared error (MSE) indicated that α ≈ 0, indicating that the LASSO method was sufficient (Hastie et al. 2009, Hooten and Hobbs 2015). Broadly speaking, all of these regularization methods for variable selection—such as LASSO and elastic net—are particularly useful in cases of strong multicollinearity among covariates (Hooten and Hobbs 2015), which are common among landscape ecology studies and which we suspected a priori. To efficiently solve the LASSO procedure and estimate model coefficients, we used the least angle regression (LAR) algorithm (Efron et al. 2004). The best value of the regulator parameter, γ1, was chosen using a modified AIC selection procedure: the best value is based on the number of steps through the LASSO path at which AIC increases after two consecutive steps (Tibshirani et al. 2017). The best model (i.e., set of non-zero coefficients) was based on the selected value of γ1. From this, P-values and 95% confidence intervals were calculated using the truncated Gaussian test described by Tibshirani et al. (2016). To conduct the LASSO, we included the following local scale covariates: in-field mean densities of Geocoris spp., fire ants, and total of the two natural enemies. We also included the factors: crop sampled for stink bugs (maize, peanut, cotton, soybean) and the year. Factors with more than two levels were split into multiple binary dummy variables. For example, we created a new binary variable for each crop species sampled; the ‘soybean’ variable was coded as a ‘1’ for each observation from a soybean field and ‘0’ for observations from the other three crop species. This was repeated for the other three crop species. The landscape covariates included the percentage area of specific crop hosts in the landscapes (maize, peanut, cotton, soybean) and total crop hosts, percentage area of GV, and the number of maize, peanut, cotton and soybean fields within 100, 500, and 1,000 m radii from the sampled field. This analysis included 32 total covariates: three natural enemy density variables, four crop species dummy variables, 3-yr dummy variables, five percentage area variables, PA ratio, percentage of GV, and 15 variables on the number of fields (of specific crops and of all crops) at different distances. Estimated densities of natural enemies were natural log transformed, and all covariates were transformed to standardized normal variables, by centering each covariate around its mean and scaling its standard deviation. For comparison, we also include the ordinary least-squares coefficient estimates, which are easily derived from LAR (Tibshirani et al. 2016). The significance level for all hypothesis tests was set at α = 0.05. All geospatial manipulations and analyses were conducted using ArcGIS for Desktop (Version 10.3, Advanced; ArcGIS for Desktop 2014). ANOVA tests were conducted in SAS (SAS Institute Inc. 1998). Randomization tests and LASSO were conducted in R 3.3.2 (R Core Team 2016). Elastic net cross-validation was conducted using the glmnet and glmnetUtils packages (Friedman et al. 2010, Ooi and Microsoft 2017); the LAR and truncated Gaussian test were conducted using the selective Inference package (Tibshirani et al. 2017). R code for running LASSO analysis can be found at https://github.com/arzeilinger/stink_bug_reproduction_lasso. Results Landscape Characteristics The percentage area of all crops combined varied over the years being higher in 2009 and 2010 than in 2011 (Supplementary Table S1, Table 1). The percentage area of maize, soybean and pasture in the landscapes were higher in 2009 than in 2010 and 2011 (Supplementary Table S1, Table 1), and the percentage area of cotton in the landscapes was lower in 2009 than in 2010 and 2011 (Supplementary Table S1, Table 1). The perimeter-to-area ratios of sampled fields, percentage area of peanut and GV in the landscape and the number of crop fields within 100 and 500 m radii from the sampled field were not significantly different over years (Supplementary Table S1, Table 1). The number of crop fields within a 1,000 m radius from the sampled field was significantly higher in 2010 than 2011 (Supplementary Table S1, Table 1). The percentage area of cotton in the landscapes (mean ± SD: 11.56 ± 5.68) was highest overall, intermediate in maize (7.16 ± 5.00) and peanut (8.03 ± 3.14) and lowest in soybean (2.00 ± 1.11). Table 1. Percentage of land use types and cropland perimeter-to-area ratio (m) for 16 landscapes within years (2009, 2010, and 2011) Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Cropland hosts = maize, peanut, cotton and soybean. GV = pastures, woodland and non-crop hosts in the woodland. Different letters within indices are significantly different at P < 0.05. View Large Table 1. Percentage of land use types and cropland perimeter-to-area ratio (m) for 16 landscapes within years (2009, 2010, and 2011) Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Cropland hosts = maize, peanut, cotton and soybean. GV = pastures, woodland and non-crop hosts in the woodland. Different letters within indices are significantly different at P < 0.05. View Large Relative Finite Rate of Population Increase Finite rates of population increase (λ generation−1) of E. servus varied significantly with crop (Table 2). Overall, λ was significantly higher in soybean than in maize, cotton and peanut (Table 3). There was high variance in λ in all of the crops, indicating that all crops were periodically low in reproduction, but sometimes they were highly productive habitats (Supplementary Table S2). Table 2. ANOVA of the effects of year, crop, and their interactions on ln-tranformed relative finite rate of population increase (λ generation−1) of E. servus. Degrees of freedom (df), Sum of Squares (SS), Mean Square (MS) Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        View Large Table 2. ANOVA of the effects of year, crop, and their interactions on ln-tranformed relative finite rate of population increase (λ generation−1) of E. servus. Degrees of freedom (df), Sum of Squares (SS), Mean Square (MS) Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        View Large Table 3. Distribution of the relative finite rate of population increase (λ generation−1) of E. servus per crop Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Means with the same letter are not significantly different at P < 0.05. View Large Table 3. Distribution of the relative finite rate of population increase (λ generation−1) of E. servus per crop Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Means with the same letter are not significantly different at P < 0.05. View Large For the LASSO analysis of E. servus λ, modified AIC selection indicated that γ1 = 9.97 produced the best model. LASSO is a method of variable selection; as such, all covariates with non-zero coefficient estimates are considered to be included in the ‘best model’. Of the 32 variables included in the analysis, only four were selected in the best model: 1) whether or not the sampled field was a soybean field, 2) mean natural enemy in-field density, 3) percentage area of cotton in the landscape, and 4) percentage area of soybean in the landscape (Table 4). All other variables were dropped from the LASSO model. Soybean field identity was the only variable with a significant truncated Gaussian test (Table 4). In contrast to the LASSO results, the least squares regression model included 29 variables, some with extremely large coefficient estimates (Table 4). Table 4. Results from LASSO analysis relating stink bug relative finite rate of population increase (λ generation−1) to a set of in-field and landscape covariates Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  aCoefficient estimates from LASSO analysis. 95% confidence interval (± 95% CI) and P-values were calculated from truncated Gaussian test. bOrdinary least-squares coefficient estimates (LS Estimates) derived from LAR output. View Large Table 4. Results from LASSO analysis relating stink bug relative finite rate of population increase (λ generation−1) to a set of in-field and landscape covariates Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  aCoefficient estimates from LASSO analysis. 95% confidence interval (± 95% CI) and P-values were calculated from truncated Gaussian test. bOrdinary least-squares coefficient estimates (LS Estimates) derived from LAR output. View Large Discussion We found that the soybean crop was the single most important predictor of increases in the relative net reproductive rate (λ generation−1) of E. servus populations, which is inconsistent with our first hypothesis. Our results also suggest that the total area of cotton and soybean in the landscape and natural enemy density in focal fields are important in affecting λ, but only in combination with each other and the soybean crop. Given that the confidence intervals for the non-significant variables overlap zero, little can be said about how they are affecting λ. Overall, we found some support that natural enemy density and cotton and soybean in the landscape have respective impacts on stink bug λ, but we were unable to isolate clear impacts of these processes on their own. In contrast to the LASSO results, the model based on least-squares regression includes 29 variables with coefficient estimates as large as ± 1014. Collinearity among covariates—which is common in spatial and landscape ecology studies—can inflate coefficient estimates in least-squares regression, as seen in our analysis (Taylor and Tibshirani 2015). LASSO and related procedures are more appropriate for such problems; the inclusion of a regulator or penalty parameter produces more conservative coefficient estimates, with more estimates equal to zero (i.e., dropped from the model) (Hooten and Hobbs 2015). The Southwood (1978) method of estimating relative net population growth rate has several potential sources of error and bias. Stink bug densities were generally low, so sampling error for a sample date was generally large. The use of areas under the incidence curve rather than incidence itself reduces bias and improves precision (Manly 1977) because the daily errors are averaged by the calculation of the area under the curve. Another source of potential bias is adult mortality and emigration from the field. If the time a colonizing adult stays in a habitat is less (or greater) than the time a progeny adult stays in the same habitat (leaving either by dispersal or death), then λ per capita will be over- (or under-) estimated. However, if such a bias in λ occurred, it would likely be a consistent bias for any particular habitat, as the factors affecting differential mortality and dispersal in the colonizing versus the progeny adults are likely to be determined by habitat factors, such as systematic variation in microclimate, nutritional quality, competition, predation, and parasitism. Thus, comparisons among landscapes within a crop habitat are likely to be accurate. If potential variation in adult mortality and emigration rates is carefully considered, differences in the estimated λ would be due to differences in net reproduction, or immature mortality as would be expected. Our results indicate that immature E. servus mortality from fire ant and Geocoris spp. predation may have reduced net reproduction across the study area. In addition, mortality from insecticides likely contributed to lower λ in landscapes with a higher proportion of cotton, as insecticides were generally applied 2–3 times on this crop. As it was not possible to obtain accurate and timely agrochemical application information from growers, we based our estimated frequency of applications on direct observations, conversations with growers and the presence in some samples of only Hippodamia convergens Guérin-Méneville (Coleoptera: Coccinellidae), a species resistant to the pyrethroids and organophosphates typically applied in Georgia (Barbosa et al. 2016). However, it was unlikely that the variance in λ among fields of a given crop was due to differential insecticide application, because cotton was similarly treated with insecticides throughout the study area, and insecticides were seldom applied on maize, peanut, and soybean. This is supported by the consistently high diversity of insect herbivores, predators and parasitoids that we observed in maize, peanut, and soybean fields (D.M.O. & J.R.R., personal observation). It is unlikely that differential emigration rates by colonizing versus progeny adult stink bugs were the major cause of variation in λ. This is because herbivorous stink bugs typically move frequently to and from nearby habitats in search of food, mates and oviposition sites (Kiritani et al.1965, Todd 1989), but they tend to remain closely associated with food plants, either foraging nearby or returning to habitats with good food plants (Kiritani et al. 1965). It is only when the food plant quality deteriorates, especially when the plant matures or is harvested, that stink bugs leave the habitat (Todd and Herzog 1980, Todd 1989, Reisig et al. 2013). We found that the higher combined densities of the generalist predators Geocoris spp. and S. invicta may have been associated with lower E. servus λ. Such a relation may have been caused by 1) density-dependent regulation of E. servus populations by generalist predators, 2) predator densities are determined by factors unrelated to E. servus population density and proportionally suppress stink bug reproduction, or 3) predator densities and E. servus reproduction co-vary because they both are responding to some common cause. While the third hypothesis cannot be ruled out, we suggest that the first hypothesis is unlikely. Given that both Geocoris spp. and S. invicta are extreme generalists, it is unlikely that stink bug densities would strongly influence predator population dynamics resulting in regulation of stink bug populations. Previous work provides some support for the second hypothesis; both predators prey on stink bug eggs (Olson and Ruberson 2012), which would reduce λ. Thus, our results are the first to suggest that predation influences stink bug reproduction in the region. A higher percentage area of either cotton or soybean in the landscape was associated with lower E. servus λ. Cotton was the dominant crop in the landscapes while soybean was the least abundant crop. Cotton may be an acceptable but not a very good reproductive host for E. servus, especially when considering the higher insecticide use on this crop. Therefore, the more cotton in the landscape, the lower would be λ in the landscape. The reasons for the negative relationship between λ and the percentage area of soybean are less clear. We found the highest E. servus λ in soybean compared to the other crops, and there was a relatively high correlation between E. servus λ and soybean suggesting that soybean can be a very good reproductive host for E. servus, as has been found for other stink bug species (Panizzi and Slansky 1985). Subsequent analyses of Geocoris spp. density indicated that soybean had strong and positive effects on their density (D. M. Olson et al. unpublished data). Higher Geocoris spp. densities and E. servus reproduction in soybean suggests that high predation on immatures may have occurred in this crop, accounting for the overall negative relationship between E. servus reproduction and the percentage area of soybean in the landscape. The stink bug N. viridula and presumably E. servus can move over distances of 1,000 m per day in search of feeding and oviposition sites (Kiritani and Sasaba 1969). However, E. servus may not need to traverse such a distance in the landscapes in the Georgia coastal plain region where crops are often closely spaced and preferred crop phenologies for feeding are present throughout the season. This is supported by the lack of any relationship between the distances of crops from the sampled field and stink bug reproduction rates. The percentage area of GV in the landscape had no relationship with λ in the sampled fields. This is contrary to the often positive relationship found for insect natural enemies, butterflies and vertebrate species (Tscharntke et al. 2012, Rusch et al. 2016), but is consistent with the relatively few studies of pest insect species (Bianchi et al. 2006, Chaplin-Kramer 2011). Woodlands in the studied landscapes were mainly comprised of natural and planted pine and oak species which likely have few nutritional resources available for stink bug species. These woodlands may have provided E. servus with summer aestivation sites, resting sites or mating sites, or temporary refuge from field disturbance and adverse abiotic conditions, (Holland and Fahrig 2000, Tscharntke et al. 2012), but these factors had minimal influence on E. servus net reproduction in the crops studied here. In a previous study, we found that E. servus was not found at the field edges of peanut, cotton and soybean that were adjacent to woodlands (Olson et al. 2012). We concluded in that study, that these woodlands were not a major source of stink bug crop colonists to peanut, cotton or soybean. This is in contrast to what Tillman and Cottrell (2016) recently found where several stink bug species, including E. servus, moved from elderberry in woodland to adjacent crops. Elderberry was not found near the crops in our study areas (Appendix A.1.1 and A.2.2 in Olson et al. 2012 ). The results from this study also suggest that the woodlands of our study were not very productive habitats for E. servus. Understanding the response of arthropod herbivores to landscape ‘complexity’ has been a focus of two recent reviews (Bianchi et al. 2006, Chaplin-Kramer et al. 2011). Bianchi et al. (2006) used the proportion of non-crop habitat and perimeter-to-area ratios and boundary density of fields as measures of complexity, and found that in 45% of the cases complexity reduced pest pressure. However, 40% of the cases showed no response to complexity. Chaplin-Kramer et al. (2011) expanded the concept of landscape complexity, and considered five measures: % natural habitat, % non-crop habitat, % crop habitat, habitat diversity (Shannon and Simpson indexes), and other measures (distance to natural habitat and length of woodland edges). They found no effect of landscape complexity on pest abundance or plant damage. Both reviews recognized that few landscape studies have measured arthropod pest responses, and Chaplin-Kramer et al. (2011) identified a need to standardize measurements of landscape complexity. They also suggest that a strong context-specific response may preclude simple standardizations. A recent study found that oviposition by two mobile, polyphagous and multivoltine moth species is dynamic and depends on the composition, arrangement, attractiveness, and preference for crops in the landscape (Parry et al. 2017). Therefore, generalist herbivores may have specific responses to a suite of factors that depend on the species and landscape context, thereby precluding simple standardization. In summary, our results showed that the landscape characteristics of the percentage area of maize, peanut and GV in the landscape and the number of crops at various distances from the sampled fields had no influence on E. servus reproduction in the landscapes. Overall, soybean was the strongest single local scale variable explaining E. servus λ. But, the combined local scale characteristics of soybean and natural enemy density and the landscape scale characteristics of the percentage area of cotton and soybean better explained E. servus λ than did soybean by itself. These results suggest that a relatively simple set of in-field and landscape variables related to differences in habitat prevalence and relative host quality influences reproduction in this mobile, polyphagous and multivoltine species. Supplementary Data Supplementary data are available at Environmental Entomology online. Acknowledgments We thank Andy Hornbuckle, Melissa Thompson, and numerous student workers for their help in the field. We also thank two anonymous reviewers for their comments which have greatly improved the manuscript. The project was supported by the National Institute of Food and Agriculture (grant number 2008-35302-04709 to D.A.A., D.M.O., and J.R.R.). References Cited Andow, D. A. 1983. The extent of monoculture and insect pest populations with particular reference to wheat and cotton. Agr. Ecosyst. Environ . 9: 25– 35. Google Scholar CrossRef Search ADS   ArcGIS for Desktop. 2014. 10.3, Advanced . Environmental Systems Research Institute, Inc., Redlands, CA. Avelino, J., A. Romero-Gurdian, H. F. Vruz-Cueller, and F. A. J. Declerck. 2012. Landscape context and scale differentially impact coffee leaf rust, coffee berry borer, and coffee root-knot nematodes. Ecol. Appl . 22: 584– 596. Google Scholar CrossRef Search ADS PubMed  Barbosa, P. R. R., J. P. Michaud, A. R. S. Rodrigues, and J. B. Torres. 2016. Dual resistance to lambda-cyhalothrin and dicrotophos in Hippodamia convergens (Coleoptera: Coccinellidae). Chemosphere . 159: 1– 9. Google Scholar CrossRef Search ADS PubMed  Blinka, E. L. 2008. Biological and ecological studies on green stink bug, Acrosternum hilare, and brown stink bug, Euschistus servus (Hemiptera: Pentatomidae), in eastern North Carolina cropping systems, PhD dissertation . North Carolina State University, Raleigh, NC. Bianchi, F. J. J. A., C. J. H. Booij, and T. Tscharntke. 2006. Sustainable pest regulation in Agricultural landscapes: a review on landscape composition: biodiversity and natural pest control. Proc. Royal Soc. B . 273: 1715– 1727. Google Scholar CrossRef Search ADS   Chaplin-Kramer, R., M. E. O’Rourke, E. J. Blitzer, and C. Kremen. 2011. A meta-analysis of crop pest and natural enemy response to landscape complexity. Ecol. Lett . 14: 922– 932. Google Scholar CrossRef Search ADS PubMed  Colunga-Garcia, M., S. H. Gage, and D. A. Landis. 1997. Response of an assemblage of Coccinellidae (Coleoptera) to a diverse agricultural landscape. Environ. Entomol . 26: 797– 804. Google Scholar CrossRef Search ADS   Efron B, T. Hastie, I. Johnstone, and R. Tibshirani. 2004. Least angle regression. Ann. Stat . 32: 407– 499. Google Scholar CrossRef Search ADS   Fabian, Y., N. Sandau, O. T. Bruggisser, A. Aebi, P. Kehrli, R. P. Rohr, R. E. Naisbit, and L. F. Bersier. 2013. The importance of landscape and spatial structure for hymenopteran-based food webs in an agro-ecosystem. J. Anim. Ecol . 82: 1203– 1214. Google Scholar CrossRef Search ADS PubMed  Friedman, J., T. Hastie, and R. Tibshirani. 2010. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw . 33: 1– 22. Google Scholar CrossRef Search ADS PubMed  Gardiner, M. M., D. A. Landis, C. Gratton, C. D. DiFonzo, M. O’Neal, J. M. Chacon, M. T. Wayo, N. P. Schmidt, E. E. Mueller, and G. E. Heimpel. 2009. Landscape diversity enhances biological control of an introduced crop pest in the north-central USA. Ecol. Appl . 19: 143– 154. Google Scholar CrossRef Search ADS PubMed  Greene, J. K., C. S. Bundy, P. M. Roberts, and B. R. Leonard. 2006. Identification and management of common boll feeding bugs in cotton . Clemson University, Blackville, SC. Hastie, T., R. Tibshirani, and J. H. Friedman. 2009. The elements of statistical learning: data mining, inference, and prediction . Springer, New York, NY. Herbert, J. J., and M. D. Toews. 2011. Seasonal abundance and population structure of brown stink bug (Hemiptera: Pentatomidae) in farmscapes containing corn, cotton, peanut, and soybean. Ann. Entomol. Soc. Am . 104: 99– 918. Google Scholar CrossRef Search ADS   Holland, J., and L. Fahrig. 2000. Effect of woody borders on insect density and diversity in crop fields: a landscape-scale analysis. Agr. Ecosyst. Environ . 78: 115– 122. Google Scholar CrossRef Search ADS   Hooten, M. B., and T. Hobbs. 2015. A guide to Bayesian model selection for ecologists. Ecol. Monogr . 85: 3– 28. Google Scholar CrossRef Search ADS   Kiritani, K., and T. Sasaba. 1969. The difference in bio- and ecological characteristics between neighbouring populations in the southern green stink bug, Nezara viridula L. Jpn. J. Ecol . 19: 177– 184. Kiritani, K., N. Hokyo, K. Kimura, and F. Nakasuji. 1965. Imaginal dispersal of the southern green stink bug Nezara viridula L., in relation to feeding and oviposition. Jpn. J. Appl. Entomol. Z . 9: 291– 297. Google Scholar CrossRef Search ADS   Koch, R. L., and T. Pahs. 2014. Species composition, abundance, and seasonal dynamics of stink bugs (Hemiptera: Pentatomidae) in Minnesota soybean fields. Environ. Entomol . 43: 883– 888. Google Scholar CrossRef Search ADS PubMed  Koch, R. L., and T. Pahs. 2015. Species composition and abundance of stink bugs (Hemiptera: Heteroptera: Pentatomidae) in Minnesota Field Corn. Environ. Entomol . 44: 233– 238. Google Scholar CrossRef Search ADS PubMed  Manly, B. F. J. 1977. A further note on Kiritani and Nakasuji’s model for stage-frequency data including comments on the use of Tukey’s jackknife technique for estimating variance. Res. Popul. Ecol . 18: 177– 186. Google Scholar CrossRef Search ADS   Marino, P. C., and D. A. Landis. 1996. Effect of landscape structure on parasitoid diversity in agroecosystems. Ecol. Appl . 6: 276– 284. Google Scholar CrossRef Search ADS   Matson, P. A., W. J. Parton, A. G. Power, and M. J. Swift. 1997. Agricultural intensification and ecosystem properties. Science . 277: 504– 509. Google Scholar CrossRef Search ADS PubMed  McPherson, J. E. 1982. The Pentatomoidea (Hemiptera) of northeastern North America with emphasis on the fauna of Illinois . Southern Illinois University Press, Carbondale and Edwardsville, IL, 240 p. McPherson, J. E., and R. M. McPherson. 2000. Stink bugs of economic importance in America North of Mexico . CRC Press, Boca Raton, FL. Google Scholar CrossRef Search ADS   Meisner, M. H., T. Zaviezo, and J. A. Rosenheim. 2017. Landscape crop composition effects on cotton yield, Lygus hesperus densities and pesticide use. Pest Manag. Sci . 73: 232– 239. Google Scholar CrossRef Search ADS PubMed  Munyaneza, J., and J. E. McPherson. 1994. Comparative-study of life-histories, laboratory rearing, and immature stages of Euschistus servus and Euschistus variolarius (Hemiptera, Pentatomidae). Great Lakes Entomol . 26: 263– 274. Naranjo, S.E., and J. L. Stimac. 1985. Development, survival, and reproduction of Geocoris punctipes (Hemiptera: Lygaeidae): effects of plant feeding on soybean and associated weeds. Environ. Entomol . 14: 523– 530. Google Scholar CrossRef Search ADS   Norton, L. R., L. G. Firbank, A. Scott, and A. R. Watkinson. 2005. Characterising spatial and temporal variation in the finite rate of population increase across the northern range boundary of the annual grass Vulpia fasciculata. Oecologia . 144: 407– 415. Google Scholar CrossRef Search ADS PubMed  Olson, D., and D. Andow. 2008. Patch edges and insect populations. Oecologia . 155: 549– 558. Google Scholar CrossRef Search ADS PubMed  Olson, D. M., and J. R. Ruberson. 2012. Crop-specific mortality of southern green stink-bug eggs in Bt- and non-Bt cotton, soybean and peanut. Biocontrol Sci. Techn . 22: 1417– 1428. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, A. R. Zeilinger, and D. A. Andow. 2011. Colonization preference of Euschistus servus and Nezara viridula in transgenic cotton varieties, peanut and soybean. Entomol. Exp. Appl . 139: 161– 169. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, and D. A. Andow. 2012. Effects on stink bugs of field edges adjacent to woodland. Agr. Ecosyst. Environ . 156: 94– 98. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, and D. A. Andow. 2016. Relative longevity of adult Nezara viridula in field cages of cotton, peanut, and soybean. Entomol. Exp. Appl . 159: 30– 36. Google Scholar CrossRef Search ADS   Ooi, H.,and Microsoft. 2017. glmnetUtils: Utilities for “Glmnet.” R package version 1.0.2. https://CRAN. R-project.org/package=glmnetUtils Panizzi, A. R., and F. Slansky. 1985. Review of phytophagous pentatomids (Hemiptera: Pentatomidae) associated with soybean in the Americas. Fla. Entomol . 68: 184– 214. Google Scholar CrossRef Search ADS   Parry, H. R., C. A. Paull, M. P. Zalucki, A. R. Ives, A. Hulthen, and N. A. Schellhorn. 2017. Estimating the landscape distribution of eggs by Helicoverpa spp., with implications for Bt resistance management. Ecol. Model . 365: 129– 140. Google Scholar CrossRef Search ADS   Pfannenstiel, R. S., and K. V. Yeargan. 1998. Association of predaceous Hemiptera with selected crops. Environ. Entomol . 27: 232– 239. Google Scholar CrossRef Search ADS   R Core Team. 2016. R: A language and environment for statistical computing . R Foundation for Statistical Computing, Vienna, Austria. Reay-Jones, F. P. 2010. Spatial and temporal patterns of stink bugs (Hemiptera: Pentatomidae) in wheat. Environ. Entomol . 39: 944– 955. Google Scholar CrossRef Search ADS PubMed  Reisig, D. D., M. Roe, and A. Dhammi. 2013. Dispersal pattern and dispersion of adult and nymph stink bugs (Hemiptera: Pentatomidae) in wheat and corn. Environ. Entomol . 42: 1184– 1192. Google Scholar CrossRef Search ADS PubMed  Rusch, A., R. Chaplin-Kramer, M. M. Gardiner, V. Hawro, J. Holland, D. Landis, C. Thies, T. Tscharntke, W. W. Weisser, C. Winqvist, et al.   2016. Agriculture landscape simplification reduces natural pest control: a quantitative synthesis. Agr. Ecosyst. Environ . 21: 198– 204. Google Scholar CrossRef Search ADS   SAS Institute Inc. 1998. SAS version 6. SAS/STAT User Guide Release, Cary, NC. Schweiger, O., J. P. Maelfait, W. van Wingerden, F. Hendrickx, R. Billeter, M. I. Speelmans, I. Augenstein, B. Aukema, S. Aviron, D. Bailey, et al.   2005. Quantifying the impact of environmental factors on arthropod communities in agricultural landscapes across organizational levels and spatial scales. J. Appl. Ecol . 42: 1129– 1139. Google Scholar CrossRef Search ADS   Sivakoff, F. S., J. A. Rosenheim, P. Dutilleul, and Y. Carrière. 2013. Influence of the surrounding landscape on crop colonization by a polyphagous insect pest. Entomol. Exp. Appl . 49: 11– 21. Google Scholar CrossRef Search ADS   Soria, M. F., P. E. Degrande, A. R. Panizzi, and M. D. Toews. 2017. Economic injury level of the Neotropical brown stink bug euschistus heros (F.) on cotton plants. Neotrop. Entomol . 46: 324– 335. Google Scholar CrossRef Search ADS PubMed  Southwood, T. R. E. 1978. The construction, description, and analysis of age-specific life-tables, pp. 356– 374. In T. R. E. Southwood (ed.), Ecological methods, with particular reference to the study of insect populations , 2nd ed. Chapman and Hall, London, United Kingdom. Taylor, J., and R. J. Tibshirani. 2015. Statistical learning and selective inference. Proc. Natl. Acad. Sci. U. S. A . 112: 7629– 7634. Google Scholar CrossRef Search ADS PubMed  Thies, C., I. Steffan-Dewenter, and T. Tscharntke. 2003. Effects of landscape context and parasitism at different spatial scales. Oikos . 101: 18– 25. Google Scholar CrossRef Search ADS   Tibshirani, R. J., J. Taylor, R. Lockhart, and R. Tibshirani. 2016. Exact post-selection inference for sequential regression procedures. J. Am. Stat. Assoc . 111: 600– 620. Google Scholar CrossRef Search ADS   Tibshirani, R. J., R. Tibshirani, J. Taylor, J. Loftus, and S. Reid. 2017. selectiveInference: Tools for Post-Selection Inference. R package version 1.2.2. https://CRAN. R-project.org/package=selectiveInference Tillman, P. G., and T. E. Cottrell. 2016. Density and egg parasitism of stink bugs (Hemiptera: Pentatomidae) in elderberry and dispersal into crops. J. Insect Sci . 16: 1– 14. Google Scholar CrossRef Search ADS PubMed  Tillman, P. G., T. D. Northfield, R. F. Mizell, and T. C. Riddle. 2009. Spatiotemporal patterns and dispersal of stink bugs (Heteroptera: Pentatomidae) in peanut-cotton farmscapes. Environ. Entomol . 38: 1038– 1052. Google Scholar CrossRef Search ADS PubMed  Tillman, P. G., T. E. Cottrell, R. F. Mizell, and E. Kramer. 2014. Effect of field edges on dispersal and distribution of colonizing stink bugs across farmscapes of the southeast USA. Bull. Entomol. Res . 104: 56– 64. Google Scholar CrossRef Search ADS PubMed  Todd, J. W. 1989. Ecology and behavior of Nezara viridula. Annu. Rev. Entomol . 34: 273– 292. Google Scholar CrossRef Search ADS   Todd, J. W., and D. C. Herzog. 1980. Sampling phytophagous Pentatomidae on soybean, pp. 438– 478. In M. Kogan and D. C. Herzog (eds.), Sampling methods in soybean entomology . Springer-Verlag, New York, NY. Google Scholar CrossRef Search ADS   Tscharntke, T., and R. Brandl. 2004. Plant-insect interactions in fragmented landscapes. Annu. Rev. Entomol . 49: 405– 430. Google Scholar CrossRef Search ADS PubMed  Tscharntke, T., R. Bommarco, Y. Clough, T. O. Crist, D. Kleijn, T. A. Rand, J. M. Tylianakis, S. Nouhuys, and S. Vidal. 2007. Conservation biological control and enemy diversity on a landscape scale. Biol. Control . 43: 294– 309. Google Scholar CrossRef Search ADS   Tscharntke, T., J. M. Tylianakis, T. A. Rand, R. K. Didham, L. Fahrig, P. Batáry, J. Bengtsson, Y. Clough, T. O. Crist, C. F. Dormann, et al.   2012. Landscape moderation of biodiversity patterns and processes - eight hypotheses. Biol. Rev. Camb. Philos. Soc . 87: 661– 685. Google Scholar CrossRef Search ADS PubMed  Turnipseed, S. G., M. J. Sullivan, J. E. Mann, and M. E. Roof. 1995. Secondary pests in transgenic Bt cotton in South Carolina, pp 768– 769. 1995 Proceedings Beltwide Cotton Conferences, San Antonio, TX, USA, January 4-7, Vol 2. Memphis: National Cotton Council. Werling, B. P., T. D. Meehan, B. A. Robertson, C. Gratton, and D. A. Landis. 2011. Biocontrol potential varies with changes in biofuel-crop plant communities and landscape perenniality. GCB Bioenergy . 3: 347– 359. Google Scholar CrossRef Search ADS   Wilson, B. P., A. J. Morris, B. E. Arroyo, S. C. Clark, and R. B. Bradbury. 1999. A review of the abundance and diversity of invertebrate and plant foods of granivorous birds in northern Europe in relation to agricultural change. Agr. Ecosyst. Environ . 65: 13– 30. Google Scholar CrossRef Search ADS   Yasuda, M., T. Mitsunaga, A. Takeda, K. Tabuchi, K. Oku, T. Yasuda, and T. Watanabe. 2011. Comparison of the effects of landscape composition on two mirid species in Japanese rice paddies. Appl. Entomol. Zool . 46: 519– 525. Google Scholar CrossRef Search ADS   Zeilinger, A. R., D. M. Olson, and D. A. Andow. 2011. Competition between stink bug and heliothine caterpillar pests on cotton at within-plant spatial scales. Entomol. Exp. Appl . 141: 59– 70. Google Scholar CrossRef Search ADS   Published by Oxford University Press on behalf of Entomological Society of America 2018. This work is written by (a) US Government employee(s) and is in the public domain in the US. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Environmental Entomology Oxford University Press

Landscape Effects on Reproduction of Euschistus servus (Hemiptera: Pentatomidae), a Mobile, Polyphagous, Multivoltine Arthropod Herbivore

Loading next page...
 
/lp/ou_press/landscape-effects-on-reproduction-of-euschistus-servus-hemiptera-IJvEZlk0Wr
Publisher
Entomological Society of America
Copyright
Published by Oxford University Press on behalf of Entomological Society of America 2018.
ISSN
0046-225X
eISSN
1938-2936
D.O.I.
10.1093/ee/nvy045
Publisher site
See Article on Publisher Site

Abstract

Abstract Landscape factors can significantly influence arthropod populations. The economically important brown stink bug, Euschistus servus (Say) (Hemiptera: Pentatomidae), is a native mobile, polyphagous and multivoltine pest of many crops in southeastern United States and understanding the relative influence of local and landscape factors on their reproduction may facilitate population management. Finite rate of population increase (λ) was estimated in four major crop hosts—maize, peanut, cotton, and soybean—over 3 yr in 16 landscapes of southern Georgia. A geographic information system (GIS) was used to characterize the surrounding landscape structure. LASSO regression was used to identify the subset of local and landscape characteristics and predator densities that account for variation in λ. The percentage area of maize, peanut and woodland and pasture in the landscape and the connectivity of cropland had no influence on E. servus λ. The best model for explaining variation in λ included only four predictor variables: whether or not the sampled field was a soybean field, mean natural enemy density in the field, percentage area of cotton in the landscape and the percentage area of soybean in the landscape. Soybean was the single most important variable for determining E. servus λ, with much greater reproduction in soybean fields than in other crop species. Penalized regression and post-selection inference provide conservative estimates of the landscape-scale determinants of E. servus reproduction and indicate that a relatively simple set of in-field and landscape variables influences reproduction in this species. brown stink bug, least angle regression There is increasing evidence that community structure, species abundance, and biotic interactions of invertebrate species in farmlands are influenced by larger-scale processes occurring at the landscape level (habitat size, spatial arrangement, connectivity and quality, and landscape matrix: Andow 1983, Marino and Landis 1996, Colunga-Garcia et al. 1997, Thies et al. 2003, Tscharntke and Brandl 2004, Schweiger et al. 2005, Bianchi et al. 2006, Tscharntke et al. 2007, Gardiner et al. 2009, Yasuda et al. 2011). Arable landscapes are often intensely managed and frequent application of agrochemicals can be a significant cause of biodiversity loss (e.g., Matson et al. 1997, Wilson et al. 1999). Further, structurally more complex landscapes—i.e., those with higher amounts of non-crop area such as woodland, hedgerows, grassland, fallows, and pastures—may compensate for locally reduced diversity inside intensively managed crop fields mainly through rapid re-colonization of resource-rich crop fields by highly dispersive organisms (Rusch et al. 2016). These non-crop areas can also provide insects with overwintering sites, summer aestivation sites, resting sites, mating sites, and sites that have spatially separated resources that are required to meet their needs (Holland and Fahrig 2000, Tscharntke et al. 2012). Additionally, spillover across habitats often increases with increasing edge density, or perimeter-area ratios, which can enhance or inhibit functional connectivity among habitats (Olson and Andow 2008, Tscharntke et al. 2012). Further, Sivakoff et al. (2013) and Meisner et al. (2017) found that crop composition immediately adjacent to a cotton field was associated with substantial differences in cotton yield, the pest species Lygus hesperus Knight (Hemiptera: Miridae) density and pesticide use, suggesting that spillover effects of arthropod species among crops of different quality may also occur in landscapes. The ability of a population to persist and increase after colonizing habitats is reflected in its finite rate of population increase (λ) in that habitat. Factors that influence λ include micro- and macro-climates, resource availability, competition, predation, and dispersal (Norton et al. 2005). Knowledge of the local and landscape variables influencing λ in different landscapes may lead to a better understanding of the factor(s) contributing to population buildup in specific areas. The southern green stink bug Nezara viridula (L.) (Hemiptera: Pentatomidae), the brown stink bug Euschistus servus (Say) (Hemiptera: Pentatomidae), and the green stink bug Chinavia hilaris (Say) (Hemiptera: Pentatomidae) are important agricultural pests in southeastern United States, that became more prominent in cotton after widespread adoption of Bt cotton and eradication of the cotton boll weevil (Turnipseed et al. 1995, Greene et al. 2006, Zeilinger et al. 2011). The major row crops economically damaged by stink bugs are field and sweet corn, soybean and cotton (McPherson and McPherson 2000; Koch and Pahs 2014, 2015; Soria et al. 2017). Although peanut—another major crop in the Southeast—is also a host of stink bugs (Tillman et al. 2009), they have not been reported to cause economic damage to this crop. In this study we concentrate on E. servus because during the last several years it has become the dominant stink bug species in southern Georgia (Olson et al. 2012). E. servus is a native species occurring from the southeastern United States west through Louisiana, Texas, New Mexico, and Arizona into California (McPherson 1982). It is bivoltine throughout its range, and overwinters in the adult stage under crop residue, leaves, pieces of bark, and in bunches of grass, preferring open fields (McPherson and McPherson 2000). This species is highly mobile and polyphagous and prefers feeding on the seeds/fruit of host plants (McPherson and McPherson 2000); thus, they move among host plants in response to changing phenology of the hosts (McPherson and McPherson 2000, Blinka 2008). Maize is an early planted host with some overlap in occurrence with later-planted peanut, cotton and soybean. All of these crops are reproductive hosts for E. servus (Herbert and Toews 2011, Koch and Pahs 2014), colonization preference for soybean is much higher than for peanut and cotton (Olson et al. 2011) and peanut is a poorer quality host in terms of adult longevity than are cotton and soybean (Olson et al. 2016). In addition, numerous non-crop E. servus hosts can exist in non-woodland and woodland field borders surrounding the crops (McPherson and McPherson 2000), and are suspected to be sources of stink bugs to the adjacent crops in the spring (Reay-Jones 2010, Olson et al. 2012, Tillman et al. 2014). Evidence for the importance of landscape-level determinants of natural enemy populations and their role in natural pest control is increasing (Werling et al. 2011, Avelino 2012, Fabian et al. 2013, Rusch et al. 2016). The role of natural enemies in stink bug population dynamics remains poorly understood, at either the field or landscape level. Olson and Ruberson (2012) indicated the importance of fire ants (Solenopsis invicta Buren, Hymenoptera: Formicidae) as predators of stink bug eggs in unsprayed cotton and peanut, whereas long-horned grasshoppers (Orthoptera: Tettigoniidae) were dominant egg predators in unsprayed soybean. Geocoris spp. (Hemiptera: Geocoridae) also feed on stink bug eggs (Olson and Ruberson 2012) and are abundant predators in cotton and soybean (Naranjo and Simac 1985, Pfannenstiel and Yeargan 1998). There is little known of the ecology of these predators in agroecosystems with respect to growth of E. servus populations. Given the variation among crop species for E. servus host quality and predation rates, a landscape approach to the study of E. servus populations encompassing major crop and non-crop hosts may reveal population patterns that can account for the present-day outbreaks in cotton and other crops. The cross-habitat spillover hypothesis put forth by Tshcarntke et al. (2012) suggests that more mobile species and species that need multiple cover types may spillover and flourish in landscapes with high functional connectivity. Therefore, we tested the hypotheses 1a) that as the percentage area of woodland and pasture, maize, peanut, cotton and/or soybean increases in the landscape, the net reproduction of E. servus increases in the landscape, 1b) as the number of maize, peanut, cotton and soybean fields closest to the focal fields increases, the net reproduction of E. servus increases in the landscape, and 2) higher densities of predators—specifically, fire ants, longhorned grasshoppers and Geocoris spp. in sampled fields reduces the net reproduction of E. servus in that field. Materials and Methods Sampling Plan We identified two areas (Southwest and East-Central) of the Coastal Plain of Georgia (Fig. 1) which to sample stink bugs in crop landscapes during the years 2009–2011. The areas were separated by approximately 150 km. Within each area, we randomly identified two or three 4.8 × 4.8 km (2,330 ha) landscapes, with two landscapes in 2011 in the Southwest area being 5.3 × 8.3 km (4,399 ha) to encompass enough of the required fields. The Southwest area had two landscape samples during 2009 (designated Shirah and Baggs), and three landscape samples during 2010 and 2011 (Vinwell, Wright and Moultrie). The East-Central area had two landscapes sampled during 2009 (North and South), and three landscape samples during 2010 (Davis, Henderson and Rufus) and 2011 (Henderson, Irwin and Rufus). Landscape samples differed in location each year because of shifting crop patterns. Working with local landowners in each landscape, we identified three fields of maize, peanut, soybean, and cotton. Each crop field was sampled weekly, from early June to August in maize, and from mid-July to late September for the other crops. The fields were commercial fields that were managed in accordance with the growers’ practices. We removed 43 fields where E. servus λ could not be estimated because of low numbers and outlier soybean fields that had a high frequency of insecticide applications by two growers in eight soybean fields (4–6 applications) or crop failure (plants less than 30 cm in height) in one grower’s three small fields (≤0.40 ha). This resulted in a total of 138 fields used in the analyses over the 3-yr period of this study. Fig. 1. View largeDownload slide A 2009 landscape within the Coastal Plain of Georgia illustrating crop types and sampled sites. Fig. 1. View largeDownload slide A 2009 landscape within the Coastal Plain of Georgia illustrating crop types and sampled sites. Each field was sampled using two permanent parallel transects (spaced 30.5 m apart) running perpendicular to the edge of the field. All sampled field edges were adjacent to woodland. A total of 20 sampling points in 2009 and 15 sampling points in 2010 and 2011 were established along each transect. The first sample point was placed 1 m from the crop edge for all years and in 2009, the next 19 samples were spaced at 5 m intervals, whereas in 2010 and 2011, samples 2 through 10 were spaced at 5 m intervals and the last five samples were spaced at 10 m intervals (=101 m from crop edge for all fields). Samples alternated weekly from crop rows 1–5 to the left and the right sides of transects to reduce repetitive plant and population disturbance. Sampling of maize was done using a two-person, whole-plant visual count with the samplers on opposite sides of the maize row for a total sampling distance of 1.5 m (ca 8 plants per sample). Peanut was sampled using a Vortis suction sampler (15 cm diameter inlet, Burkard Manufacturing Company, Ltd., Hertfordshire, UK) with 12 × 3-s suctions at 7–8 cm from the soil at each sampling point (2,121 cm2 sampled per sampling station). Cotton and soybean were sampled using a 1.5 m drop cloth with 10 shakes per sample. In all crops, about 1.5 linear row-meters of plants were sampled at each point. All stink bugs were identified and assigned to a developmental stage, but only E. servus was common enough over the 3 yr to analyze in detail. Landscape Characteristics For each landscape, we used a geographic information system (GIS) to determine the percentage of area in the landscape that contained E. servus host crops (intensively produced maize, peanut, cotton and soybean); semi-natural habitat consisting of woodland and pasture or ‘green-veining’ (GV), and the number of maize, peanut, cotton and soybean fields within 100, 500, and 1,000 m radii from the focal sampling field as a measure of cropland connectivity. Assuming no physical barriers to E. servus movement, we calculated straight-line connections from the edges of sampled crop fields to the closest edges of nearby crops. Females of an ecologically similar species, N. viridula, have been observed dispersing up to 1,000 m per day by flight in search of feeding or oviposition sites (Kiritani and Sasaba 1969). Natural Enemies We included S. invicta, adults and nymphs of longhorned grasshoppers, and adults and nymphs of Geocoris spp. in our weekly samples. In addition to generalist predator sampling, we estimated parasitism rates by collecting and incubating a subsample of stink bug adults and nymphs. Numbers of natural enemies and parasitism rates were averaged over field samples to coincide with our estimates of λ per field. However, because parasitism of adults (1.4%) and nymphs (0%) was very low, overall parasitism was excluded from subsequent analysis. Longhorned grasshopper density was very low so they were also excluded from subsequent analysis. Relative Finite Rate of Population Increase Relative finite rate of population increase (λ generation−1) was estimated for each sampled field where adults were observed in the field. The date that adults were first observed was considered to be the date of first colonization and the dates that peaks in adult density occurred were noted. If late-instar nymphs were observed before adults were observed, the date of first adult colonization was assumed to have occurred 20 d before the date of first sighting of nymphs, based on E. servus developmental time (Munyaneza & McPherson 1994). On average over all crops, the number of days from the first adult peak to the second adult peak was 39.7 ± 2.5 d, so during the 40 d following the date of first adult colonization, any adults observed were considered to be colonists. After this period, adults were considered to be progeny of the colonists. Eggs were too infrequently sampled to be used to estimate an oviposition period or an initial density of the offspring generation, and nymphs were found less frequently than adults. Thus, we used colonizing and progeny adults to estimate λ. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). An example is shown in Fig. 2. Fig. 2. View largeDownload slide An example of incidence data: the average number of brown stink bug adults, nymphs, and eggs per sample versus day of year in maize in the North Coffee landscape of the East-Central region. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). Fig. 2. View largeDownload slide An example of incidence data: the average number of brown stink bug adults, nymphs, and eggs per sample versus day of year in maize in the North Coffee landscape of the East-Central region. Using Southwood’s method (Southwood 1978), we calculated the area under the colonist-incidence curve (Ac) and the progeny-incidence curve (Ap), and estimated relative net population growth as the ratio of these two (Ap/Ac). Statistical Analyses First, we investigated cropland and non-cropland heterogeneity over years. We analyzed the effects of year on arcsine-square root transformed percentage area of cropland hosts, GV (woodland and pasture), as well as the perimeter-to-area ratio of the sampled fields, and the number of cropland fields within 100, 500, and 1,000 m from the sampled field with ANOVA and used Tukey’s HSD to separate the means. Second, we investigated which factors explained E. servus population increase, λ. We included three factors in this analysis: 3 yr, four or six landscapes (landscapes varied over years), and three maize, peanut, cotton and soybean fields in each landscape. During 2009, four landscapes were sampled and during 2010 and 2011 six landscapes were sampled. The design was a nested ANOVA with years as whole plot factors and whole plot error defined by landscapes within years (error 1). The subplots were the crops (and crop interactions with year), and groups were the unit of replication (error 2) after removing all landscape interactions with crops (and landscape by crop interactions with year). The natural logarithm of λ (lnλ) was analyzed as the untransformed λ was highly skewed and heteroscedastic. The standardized residuals for error 2 were normal (Supplementary Fig. S1a and b), and there was no correlation between the standardized residuals and the predicted values (Supplementary Fig. S2) thereby, satisfying assumptions of an ANOVA. However, lnλ error was still slightly heteroscedastic, so we also conducted a randomization test, using the same ANOVA model. Values of lnλ were randomized among crops and groups within landscapes, and landscapes were randomized among years, making sure that the unbalanced structure was preserved. These randomizations independently permuted the units of replication with respect to year at the whole plot level and with respect to crop (and its interactions) at the subplot level. The estimated F-value from each randomization for each term of interest was used to construct null distributions, and the observed F was used to calculate the P-value. The simulations were repeated 2.5 million times so that the estimated P-value was accurate to five significant digits. Third, we determined the set of landscape and local scale variables that best explained variation in E. servus reproduction using the LASSO method of variable selection (Hastie, Tibshirani, and Friedman 2009). Stepwise regression and model averaging are the other two ways to select a subset of variables. Stepwise regression generally overestimates parameter values, and underestimates standard errors. Model averaging requires a priori specification of a set of models. The LASSO procedure uses a regularization penalty to ‘shrink’ coefficient estimates of a linear combination of covariates, using a pre-determined estimate of the regulator or penalty parameter, γ1 (note, while other authors have represented the LASSO regulator as λ, we adopt the notation of Hooten and Hobbs (2015) to differentiate it from our estimates of stink bug reproduction, λ). The LASSO procedure is a special case of the elastic net method (Hastie et al. 2009); we conducted K-fold cross-validation of the elastic net tuning parameter, α (results not shown) the minimum mean-squared error (MSE) indicated that α ≈ 0, indicating that the LASSO method was sufficient (Hastie et al. 2009, Hooten and Hobbs 2015). Broadly speaking, all of these regularization methods for variable selection—such as LASSO and elastic net—are particularly useful in cases of strong multicollinearity among covariates (Hooten and Hobbs 2015), which are common among landscape ecology studies and which we suspected a priori. To efficiently solve the LASSO procedure and estimate model coefficients, we used the least angle regression (LAR) algorithm (Efron et al. 2004). The best value of the regulator parameter, γ1, was chosen using a modified AIC selection procedure: the best value is based on the number of steps through the LASSO path at which AIC increases after two consecutive steps (Tibshirani et al. 2017). The best model (i.e., set of non-zero coefficients) was based on the selected value of γ1. From this, P-values and 95% confidence intervals were calculated using the truncated Gaussian test described by Tibshirani et al. (2016). To conduct the LASSO, we included the following local scale covariates: in-field mean densities of Geocoris spp., fire ants, and total of the two natural enemies. We also included the factors: crop sampled for stink bugs (maize, peanut, cotton, soybean) and the year. Factors with more than two levels were split into multiple binary dummy variables. For example, we created a new binary variable for each crop species sampled; the ‘soybean’ variable was coded as a ‘1’ for each observation from a soybean field and ‘0’ for observations from the other three crop species. This was repeated for the other three crop species. The landscape covariates included the percentage area of specific crop hosts in the landscapes (maize, peanut, cotton, soybean) and total crop hosts, percentage area of GV, and the number of maize, peanut, cotton and soybean fields within 100, 500, and 1,000 m radii from the sampled field. This analysis included 32 total covariates: three natural enemy density variables, four crop species dummy variables, 3-yr dummy variables, five percentage area variables, PA ratio, percentage of GV, and 15 variables on the number of fields (of specific crops and of all crops) at different distances. Estimated densities of natural enemies were natural log transformed, and all covariates were transformed to standardized normal variables, by centering each covariate around its mean and scaling its standard deviation. For comparison, we also include the ordinary least-squares coefficient estimates, which are easily derived from LAR (Tibshirani et al. 2016). The significance level for all hypothesis tests was set at α = 0.05. All geospatial manipulations and analyses were conducted using ArcGIS for Desktop (Version 10.3, Advanced; ArcGIS for Desktop 2014). ANOVA tests were conducted in SAS (SAS Institute Inc. 1998). Randomization tests and LASSO were conducted in R 3.3.2 (R Core Team 2016). Elastic net cross-validation was conducted using the glmnet and glmnetUtils packages (Friedman et al. 2010, Ooi and Microsoft 2017); the LAR and truncated Gaussian test were conducted using the selective Inference package (Tibshirani et al. 2017). R code for running LASSO analysis can be found at https://github.com/arzeilinger/stink_bug_reproduction_lasso. Results Landscape Characteristics The percentage area of all crops combined varied over the years being higher in 2009 and 2010 than in 2011 (Supplementary Table S1, Table 1). The percentage area of maize, soybean and pasture in the landscapes were higher in 2009 than in 2010 and 2011 (Supplementary Table S1, Table 1), and the percentage area of cotton in the landscapes was lower in 2009 than in 2010 and 2011 (Supplementary Table S1, Table 1). The perimeter-to-area ratios of sampled fields, percentage area of peanut and GV in the landscape and the number of crop fields within 100 and 500 m radii from the sampled field were not significantly different over years (Supplementary Table S1, Table 1). The number of crop fields within a 1,000 m radius from the sampled field was significantly higher in 2010 than 2011 (Supplementary Table S1, Table 1). The percentage area of cotton in the landscapes (mean ± SD: 11.56 ± 5.68) was highest overall, intermediate in maize (7.16 ± 5.00) and peanut (8.03 ± 3.14) and lowest in soybean (2.00 ± 1.11). Table 1. Percentage of land use types and cropland perimeter-to-area ratio (m) for 16 landscapes within years (2009, 2010, and 2011) Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Cropland hosts = maize, peanut, cotton and soybean. GV = pastures, woodland and non-crop hosts in the woodland. Different letters within indices are significantly different at P < 0.05. View Large Table 1. Percentage of land use types and cropland perimeter-to-area ratio (m) for 16 landscapes within years (2009, 2010, and 2011) Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Index  Year  Mean ± SE  % Cropland hosts  2009  30.65 ± 0.020a  2010  29.66 ± 0.007a  2011  25.57 ± 0.008b  % Maize  2009  0.101 ± 0.009a  2010  0.067 ± 0.006b  2011  0.057 ± 0.004b  % Peanut  2009  0.073 ± 0.006  2010  0.077 ± 0.004  2011  0.064 ± 0.004  % Cotton  2009  0.082 ± 0.003b  2010  0.132 ± 0.006a  2011  0.124 ± 0.009a  % Soybean  2009  0.052 ± 0.004a  2010  0.022 ± 0.001b  2011  0.010 ± 0.001b  % Green-veining  2009  0.421 ± 0.019  2010  0.390 ± 0.013  2011  0.405 ± 0.009  Perimeter-area-ratio  2009  136.72 ± 13.72  2010  134.52 ± 8.59  2011  119.09 ± 7.71  Number of crop fields (100 m)  2009  4.00 ± 0.30  2010  3.39 ± 0.20  2011  3.83 ± 0.20  Number of crop fields (500 m)  2009  9.83 ± 0.64  2010  9.48 ± 0.54  2011  8.00 ± 0.49  Number of crop fields (1,000 m)  2009  18.74 ± 1.10ab  2010  20.04 ± 0.98a  2011  16.00 ± 0.67b  Cropland hosts = maize, peanut, cotton and soybean. GV = pastures, woodland and non-crop hosts in the woodland. Different letters within indices are significantly different at P < 0.05. View Large Relative Finite Rate of Population Increase Finite rates of population increase (λ generation−1) of E. servus varied significantly with crop (Table 2). Overall, λ was significantly higher in soybean than in maize, cotton and peanut (Table 3). There was high variance in λ in all of the crops, indicating that all crops were periodically low in reproduction, but sometimes they were highly productive habitats (Supplementary Table S2). Table 2. ANOVA of the effects of year, crop, and their interactions on ln-tranformed relative finite rate of population increase (λ generation−1) of E. servus. Degrees of freedom (df), Sum of Squares (SS), Mean Square (MS) Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        View Large Table 2. ANOVA of the effects of year, crop, and their interactions on ln-tranformed relative finite rate of population increase (λ generation−1) of E. servus. Degrees of freedom (df), Sum of Squares (SS), Mean Square (MS) Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        Source  df  Type III SS  MS  F-value  P-value GLM  P-value randomization  Level 1   Year  2  0.81078484  0.40539242  0.57  0.560  0.989   Error 1  5  6.64969660  2.2223104        Level 2   Crop  3  7.51726378  1.67410628  3.50  0.018  <0.001   Year × crop  6  6.36069670  1.06011612  1.48  0.193  0.408   Error 2  99  70.9751941  0.7169212        View Large Table 3. Distribution of the relative finite rate of population increase (λ generation−1) of E. servus per crop Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Means with the same letter are not significantly different at P < 0.05. View Large Table 3. Distribution of the relative finite rate of population increase (λ generation−1) of E. servus per crop Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Level of crop  n  λ Generation−1  Mean  Standard error  Maize  41  1.33b  0.15  Cotton  28  1.13b  0.14  Peanut  36  1.33b  0.29  Soybean  24  3.96a  0.63  Means with the same letter are not significantly different at P < 0.05. View Large For the LASSO analysis of E. servus λ, modified AIC selection indicated that γ1 = 9.97 produced the best model. LASSO is a method of variable selection; as such, all covariates with non-zero coefficient estimates are considered to be included in the ‘best model’. Of the 32 variables included in the analysis, only four were selected in the best model: 1) whether or not the sampled field was a soybean field, 2) mean natural enemy in-field density, 3) percentage area of cotton in the landscape, and 4) percentage area of soybean in the landscape (Table 4). All other variables were dropped from the LASSO model. Soybean field identity was the only variable with a significant truncated Gaussian test (Table 4). In contrast to the LASSO results, the least squares regression model included 29 variables, some with extremely large coefficient estimates (Table 4). Table 4. Results from LASSO analysis relating stink bug relative finite rate of population increase (λ generation−1) to a set of in-field and landscape covariates Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  aCoefficient estimates from LASSO analysis. 95% confidence interval (± 95% CI) and P-values were calculated from truncated Gaussian test. bOrdinary least-squares coefficient estimates (LS Estimates) derived from LAR output. View Large Table 4. Results from LASSO analysis relating stink bug relative finite rate of population increase (λ generation−1) to a set of in-field and landscape covariates Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  Effect  LASSO estimate (± 95% CI)a  P-valuea  LS estimateb  Soybean  0.338 (0.203, 1.13)  0.000746  0.661  Mean fire ant and Geocoris spp.  −0.129 (−0.484, 4.88)  0.822  0.0366  Percentage of cotton  −0.0477 (−1.62, 1.38)  0.417  −2.63E^14  Percentage of soybean  −0.0183 (−1.95, 3.34)  0.603  −9.59E^13  GV  0  NA  0.498  PA  0  NA  −0.252  Mean Geocoris spp.  0  NA  −0.177  Mean ants  0  NA  −0.172  Percentage of maize  0  NA  −2.20E^14  Percentage of peanut  0  NA  −1.77E^14  Percentage of all crops  0  NA  3.95E^14  Number of maize (100 m)  0  NA  −1.07E^13  Number of cotton (100 m)  0  NA  −1.47E^13  Number of peanut (100 m)  0  NA  −1.27E^13  Number of soybean (100 m)  0  NA  −1.16E^13  Number of all crops (100 m)  0  NA  1.99E^13  Number of maize (500 m)  0  NA  0.158  Number of cotton (500 m)  0  NA  −0.324  Number of peanut (500 m)  0  NA  0.0509  Number of soybean (500 m)  0  NA  0.289  Number all crops (500 m)  0  NA  0  Number of maize (1,000 m)  0  NA  −0.187  Number of cotton (1,000 m)  0  NA  0.233  Number of peanut (1,000 m)  0  NA  0.184  Number of soybean (1,000 m)  0  NA  −0.0777  Number of all crops (1,000 m)  0  NA  0  Maize  0  NA  0.119  Cotton  0  NA  0  Peanut  0  NA  0.012  Year 2009  0  NA  1.85E^14  Year 2010  0  NA  2.12E^14  Year 2011  0  NA  2.05E^14  aCoefficient estimates from LASSO analysis. 95% confidence interval (± 95% CI) and P-values were calculated from truncated Gaussian test. bOrdinary least-squares coefficient estimates (LS Estimates) derived from LAR output. View Large Discussion We found that the soybean crop was the single most important predictor of increases in the relative net reproductive rate (λ generation−1) of E. servus populations, which is inconsistent with our first hypothesis. Our results also suggest that the total area of cotton and soybean in the landscape and natural enemy density in focal fields are important in affecting λ, but only in combination with each other and the soybean crop. Given that the confidence intervals for the non-significant variables overlap zero, little can be said about how they are affecting λ. Overall, we found some support that natural enemy density and cotton and soybean in the landscape have respective impacts on stink bug λ, but we were unable to isolate clear impacts of these processes on their own. In contrast to the LASSO results, the model based on least-squares regression includes 29 variables with coefficient estimates as large as ± 1014. Collinearity among covariates—which is common in spatial and landscape ecology studies—can inflate coefficient estimates in least-squares regression, as seen in our analysis (Taylor and Tibshirani 2015). LASSO and related procedures are more appropriate for such problems; the inclusion of a regulator or penalty parameter produces more conservative coefficient estimates, with more estimates equal to zero (i.e., dropped from the model) (Hooten and Hobbs 2015). The Southwood (1978) method of estimating relative net population growth rate has several potential sources of error and bias. Stink bug densities were generally low, so sampling error for a sample date was generally large. The use of areas under the incidence curve rather than incidence itself reduces bias and improves precision (Manly 1977) because the daily errors are averaged by the calculation of the area under the curve. Another source of potential bias is adult mortality and emigration from the field. If the time a colonizing adult stays in a habitat is less (or greater) than the time a progeny adult stays in the same habitat (leaving either by dispersal or death), then λ per capita will be over- (or under-) estimated. However, if such a bias in λ occurred, it would likely be a consistent bias for any particular habitat, as the factors affecting differential mortality and dispersal in the colonizing versus the progeny adults are likely to be determined by habitat factors, such as systematic variation in microclimate, nutritional quality, competition, predation, and parasitism. Thus, comparisons among landscapes within a crop habitat are likely to be accurate. If potential variation in adult mortality and emigration rates is carefully considered, differences in the estimated λ would be due to differences in net reproduction, or immature mortality as would be expected. Our results indicate that immature E. servus mortality from fire ant and Geocoris spp. predation may have reduced net reproduction across the study area. In addition, mortality from insecticides likely contributed to lower λ in landscapes with a higher proportion of cotton, as insecticides were generally applied 2–3 times on this crop. As it was not possible to obtain accurate and timely agrochemical application information from growers, we based our estimated frequency of applications on direct observations, conversations with growers and the presence in some samples of only Hippodamia convergens Guérin-Méneville (Coleoptera: Coccinellidae), a species resistant to the pyrethroids and organophosphates typically applied in Georgia (Barbosa et al. 2016). However, it was unlikely that the variance in λ among fields of a given crop was due to differential insecticide application, because cotton was similarly treated with insecticides throughout the study area, and insecticides were seldom applied on maize, peanut, and soybean. This is supported by the consistently high diversity of insect herbivores, predators and parasitoids that we observed in maize, peanut, and soybean fields (D.M.O. & J.R.R., personal observation). It is unlikely that differential emigration rates by colonizing versus progeny adult stink bugs were the major cause of variation in λ. This is because herbivorous stink bugs typically move frequently to and from nearby habitats in search of food, mates and oviposition sites (Kiritani et al.1965, Todd 1989), but they tend to remain closely associated with food plants, either foraging nearby or returning to habitats with good food plants (Kiritani et al. 1965). It is only when the food plant quality deteriorates, especially when the plant matures or is harvested, that stink bugs leave the habitat (Todd and Herzog 1980, Todd 1989, Reisig et al. 2013). We found that the higher combined densities of the generalist predators Geocoris spp. and S. invicta may have been associated with lower E. servus λ. Such a relation may have been caused by 1) density-dependent regulation of E. servus populations by generalist predators, 2) predator densities are determined by factors unrelated to E. servus population density and proportionally suppress stink bug reproduction, or 3) predator densities and E. servus reproduction co-vary because they both are responding to some common cause. While the third hypothesis cannot be ruled out, we suggest that the first hypothesis is unlikely. Given that both Geocoris spp. and S. invicta are extreme generalists, it is unlikely that stink bug densities would strongly influence predator population dynamics resulting in regulation of stink bug populations. Previous work provides some support for the second hypothesis; both predators prey on stink bug eggs (Olson and Ruberson 2012), which would reduce λ. Thus, our results are the first to suggest that predation influences stink bug reproduction in the region. A higher percentage area of either cotton or soybean in the landscape was associated with lower E. servus λ. Cotton was the dominant crop in the landscapes while soybean was the least abundant crop. Cotton may be an acceptable but not a very good reproductive host for E. servus, especially when considering the higher insecticide use on this crop. Therefore, the more cotton in the landscape, the lower would be λ in the landscape. The reasons for the negative relationship between λ and the percentage area of soybean are less clear. We found the highest E. servus λ in soybean compared to the other crops, and there was a relatively high correlation between E. servus λ and soybean suggesting that soybean can be a very good reproductive host for E. servus, as has been found for other stink bug species (Panizzi and Slansky 1985). Subsequent analyses of Geocoris spp. density indicated that soybean had strong and positive effects on their density (D. M. Olson et al. unpublished data). Higher Geocoris spp. densities and E. servus reproduction in soybean suggests that high predation on immatures may have occurred in this crop, accounting for the overall negative relationship between E. servus reproduction and the percentage area of soybean in the landscape. The stink bug N. viridula and presumably E. servus can move over distances of 1,000 m per day in search of feeding and oviposition sites (Kiritani and Sasaba 1969). However, E. servus may not need to traverse such a distance in the landscapes in the Georgia coastal plain region where crops are often closely spaced and preferred crop phenologies for feeding are present throughout the season. This is supported by the lack of any relationship between the distances of crops from the sampled field and stink bug reproduction rates. The percentage area of GV in the landscape had no relationship with λ in the sampled fields. This is contrary to the often positive relationship found for insect natural enemies, butterflies and vertebrate species (Tscharntke et al. 2012, Rusch et al. 2016), but is consistent with the relatively few studies of pest insect species (Bianchi et al. 2006, Chaplin-Kramer 2011). Woodlands in the studied landscapes were mainly comprised of natural and planted pine and oak species which likely have few nutritional resources available for stink bug species. These woodlands may have provided E. servus with summer aestivation sites, resting sites or mating sites, or temporary refuge from field disturbance and adverse abiotic conditions, (Holland and Fahrig 2000, Tscharntke et al. 2012), but these factors had minimal influence on E. servus net reproduction in the crops studied here. In a previous study, we found that E. servus was not found at the field edges of peanut, cotton and soybean that were adjacent to woodlands (Olson et al. 2012). We concluded in that study, that these woodlands were not a major source of stink bug crop colonists to peanut, cotton or soybean. This is in contrast to what Tillman and Cottrell (2016) recently found where several stink bug species, including E. servus, moved from elderberry in woodland to adjacent crops. Elderberry was not found near the crops in our study areas (Appendix A.1.1 and A.2.2 in Olson et al. 2012 ). The results from this study also suggest that the woodlands of our study were not very productive habitats for E. servus. Understanding the response of arthropod herbivores to landscape ‘complexity’ has been a focus of two recent reviews (Bianchi et al. 2006, Chaplin-Kramer et al. 2011). Bianchi et al. (2006) used the proportion of non-crop habitat and perimeter-to-area ratios and boundary density of fields as measures of complexity, and found that in 45% of the cases complexity reduced pest pressure. However, 40% of the cases showed no response to complexity. Chaplin-Kramer et al. (2011) expanded the concept of landscape complexity, and considered five measures: % natural habitat, % non-crop habitat, % crop habitat, habitat diversity (Shannon and Simpson indexes), and other measures (distance to natural habitat and length of woodland edges). They found no effect of landscape complexity on pest abundance or plant damage. Both reviews recognized that few landscape studies have measured arthropod pest responses, and Chaplin-Kramer et al. (2011) identified a need to standardize measurements of landscape complexity. They also suggest that a strong context-specific response may preclude simple standardizations. A recent study found that oviposition by two mobile, polyphagous and multivoltine moth species is dynamic and depends on the composition, arrangement, attractiveness, and preference for crops in the landscape (Parry et al. 2017). Therefore, generalist herbivores may have specific responses to a suite of factors that depend on the species and landscape context, thereby precluding simple standardization. In summary, our results showed that the landscape characteristics of the percentage area of maize, peanut and GV in the landscape and the number of crops at various distances from the sampled fields had no influence on E. servus reproduction in the landscapes. Overall, soybean was the strongest single local scale variable explaining E. servus λ. But, the combined local scale characteristics of soybean and natural enemy density and the landscape scale characteristics of the percentage area of cotton and soybean better explained E. servus λ than did soybean by itself. These results suggest that a relatively simple set of in-field and landscape variables related to differences in habitat prevalence and relative host quality influences reproduction in this mobile, polyphagous and multivoltine species. Supplementary Data Supplementary data are available at Environmental Entomology online. Acknowledgments We thank Andy Hornbuckle, Melissa Thompson, and numerous student workers for their help in the field. We also thank two anonymous reviewers for their comments which have greatly improved the manuscript. The project was supported by the National Institute of Food and Agriculture (grant number 2008-35302-04709 to D.A.A., D.M.O., and J.R.R.). References Cited Andow, D. A. 1983. The extent of monoculture and insect pest populations with particular reference to wheat and cotton. Agr. Ecosyst. Environ . 9: 25– 35. Google Scholar CrossRef Search ADS   ArcGIS for Desktop. 2014. 10.3, Advanced . Environmental Systems Research Institute, Inc., Redlands, CA. Avelino, J., A. Romero-Gurdian, H. F. Vruz-Cueller, and F. A. J. Declerck. 2012. Landscape context and scale differentially impact coffee leaf rust, coffee berry borer, and coffee root-knot nematodes. Ecol. Appl . 22: 584– 596. Google Scholar CrossRef Search ADS PubMed  Barbosa, P. R. R., J. P. Michaud, A. R. S. Rodrigues, and J. B. Torres. 2016. Dual resistance to lambda-cyhalothrin and dicrotophos in Hippodamia convergens (Coleoptera: Coccinellidae). Chemosphere . 159: 1– 9. Google Scholar CrossRef Search ADS PubMed  Blinka, E. L. 2008. Biological and ecological studies on green stink bug, Acrosternum hilare, and brown stink bug, Euschistus servus (Hemiptera: Pentatomidae), in eastern North Carolina cropping systems, PhD dissertation . North Carolina State University, Raleigh, NC. Bianchi, F. J. J. A., C. J. H. Booij, and T. Tscharntke. 2006. Sustainable pest regulation in Agricultural landscapes: a review on landscape composition: biodiversity and natural pest control. Proc. Royal Soc. B . 273: 1715– 1727. Google Scholar CrossRef Search ADS   Chaplin-Kramer, R., M. E. O’Rourke, E. J. Blitzer, and C. Kremen. 2011. A meta-analysis of crop pest and natural enemy response to landscape complexity. Ecol. Lett . 14: 922– 932. Google Scholar CrossRef Search ADS PubMed  Colunga-Garcia, M., S. H. Gage, and D. A. Landis. 1997. Response of an assemblage of Coccinellidae (Coleoptera) to a diverse agricultural landscape. Environ. Entomol . 26: 797– 804. Google Scholar CrossRef Search ADS   Efron B, T. Hastie, I. Johnstone, and R. Tibshirani. 2004. Least angle regression. Ann. Stat . 32: 407– 499. Google Scholar CrossRef Search ADS   Fabian, Y., N. Sandau, O. T. Bruggisser, A. Aebi, P. Kehrli, R. P. Rohr, R. E. Naisbit, and L. F. Bersier. 2013. The importance of landscape and spatial structure for hymenopteran-based food webs in an agro-ecosystem. J. Anim. Ecol . 82: 1203– 1214. Google Scholar CrossRef Search ADS PubMed  Friedman, J., T. Hastie, and R. Tibshirani. 2010. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw . 33: 1– 22. Google Scholar CrossRef Search ADS PubMed  Gardiner, M. M., D. A. Landis, C. Gratton, C. D. DiFonzo, M. O’Neal, J. M. Chacon, M. T. Wayo, N. P. Schmidt, E. E. Mueller, and G. E. Heimpel. 2009. Landscape diversity enhances biological control of an introduced crop pest in the north-central USA. Ecol. Appl . 19: 143– 154. Google Scholar CrossRef Search ADS PubMed  Greene, J. K., C. S. Bundy, P. M. Roberts, and B. R. Leonard. 2006. Identification and management of common boll feeding bugs in cotton . Clemson University, Blackville, SC. Hastie, T., R. Tibshirani, and J. H. Friedman. 2009. The elements of statistical learning: data mining, inference, and prediction . Springer, New York, NY. Herbert, J. J., and M. D. Toews. 2011. Seasonal abundance and population structure of brown stink bug (Hemiptera: Pentatomidae) in farmscapes containing corn, cotton, peanut, and soybean. Ann. Entomol. Soc. Am . 104: 99– 918. Google Scholar CrossRef Search ADS   Holland, J., and L. Fahrig. 2000. Effect of woody borders on insect density and diversity in crop fields: a landscape-scale analysis. Agr. Ecosyst. Environ . 78: 115– 122. Google Scholar CrossRef Search ADS   Hooten, M. B., and T. Hobbs. 2015. A guide to Bayesian model selection for ecologists. Ecol. Monogr . 85: 3– 28. Google Scholar CrossRef Search ADS   Kiritani, K., and T. Sasaba. 1969. The difference in bio- and ecological characteristics between neighbouring populations in the southern green stink bug, Nezara viridula L. Jpn. J. Ecol . 19: 177– 184. Kiritani, K., N. Hokyo, K. Kimura, and F. Nakasuji. 1965. Imaginal dispersal of the southern green stink bug Nezara viridula L., in relation to feeding and oviposition. Jpn. J. Appl. Entomol. Z . 9: 291– 297. Google Scholar CrossRef Search ADS   Koch, R. L., and T. Pahs. 2014. Species composition, abundance, and seasonal dynamics of stink bugs (Hemiptera: Pentatomidae) in Minnesota soybean fields. Environ. Entomol . 43: 883– 888. Google Scholar CrossRef Search ADS PubMed  Koch, R. L., and T. Pahs. 2015. Species composition and abundance of stink bugs (Hemiptera: Heteroptera: Pentatomidae) in Minnesota Field Corn. Environ. Entomol . 44: 233– 238. Google Scholar CrossRef Search ADS PubMed  Manly, B. F. J. 1977. A further note on Kiritani and Nakasuji’s model for stage-frequency data including comments on the use of Tukey’s jackknife technique for estimating variance. Res. Popul. Ecol . 18: 177– 186. Google Scholar CrossRef Search ADS   Marino, P. C., and D. A. Landis. 1996. Effect of landscape structure on parasitoid diversity in agroecosystems. Ecol. Appl . 6: 276– 284. Google Scholar CrossRef Search ADS   Matson, P. A., W. J. Parton, A. G. Power, and M. J. Swift. 1997. Agricultural intensification and ecosystem properties. Science . 277: 504– 509. Google Scholar CrossRef Search ADS PubMed  McPherson, J. E. 1982. The Pentatomoidea (Hemiptera) of northeastern North America with emphasis on the fauna of Illinois . Southern Illinois University Press, Carbondale and Edwardsville, IL, 240 p. McPherson, J. E., and R. M. McPherson. 2000. Stink bugs of economic importance in America North of Mexico . CRC Press, Boca Raton, FL. Google Scholar CrossRef Search ADS   Meisner, M. H., T. Zaviezo, and J. A. Rosenheim. 2017. Landscape crop composition effects on cotton yield, Lygus hesperus densities and pesticide use. Pest Manag. Sci . 73: 232– 239. Google Scholar CrossRef Search ADS PubMed  Munyaneza, J., and J. E. McPherson. 1994. Comparative-study of life-histories, laboratory rearing, and immature stages of Euschistus servus and Euschistus variolarius (Hemiptera, Pentatomidae). Great Lakes Entomol . 26: 263– 274. Naranjo, S.E., and J. L. Stimac. 1985. Development, survival, and reproduction of Geocoris punctipes (Hemiptera: Lygaeidae): effects of plant feeding on soybean and associated weeds. Environ. Entomol . 14: 523– 530. Google Scholar CrossRef Search ADS   Norton, L. R., L. G. Firbank, A. Scott, and A. R. Watkinson. 2005. Characterising spatial and temporal variation in the finite rate of population increase across the northern range boundary of the annual grass Vulpia fasciculata. Oecologia . 144: 407– 415. Google Scholar CrossRef Search ADS PubMed  Olson, D., and D. Andow. 2008. Patch edges and insect populations. Oecologia . 155: 549– 558. Google Scholar CrossRef Search ADS PubMed  Olson, D. M., and J. R. Ruberson. 2012. Crop-specific mortality of southern green stink-bug eggs in Bt- and non-Bt cotton, soybean and peanut. Biocontrol Sci. Techn . 22: 1417– 1428. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, A. R. Zeilinger, and D. A. Andow. 2011. Colonization preference of Euschistus servus and Nezara viridula in transgenic cotton varieties, peanut and soybean. Entomol. Exp. Appl . 139: 161– 169. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, and D. A. Andow. 2012. Effects on stink bugs of field edges adjacent to woodland. Agr. Ecosyst. Environ . 156: 94– 98. Google Scholar CrossRef Search ADS   Olson, D. M., J. R. Ruberson, and D. A. Andow. 2016. Relative longevity of adult Nezara viridula in field cages of cotton, peanut, and soybean. Entomol. Exp. Appl . 159: 30– 36. Google Scholar CrossRef Search ADS   Ooi, H.,and Microsoft. 2017. glmnetUtils: Utilities for “Glmnet.” R package version 1.0.2. https://CRAN. R-project.org/package=glmnetUtils Panizzi, A. R., and F. Slansky. 1985. Review of phytophagous pentatomids (Hemiptera: Pentatomidae) associated with soybean in the Americas. Fla. Entomol . 68: 184– 214. Google Scholar CrossRef Search ADS   Parry, H. R., C. A. Paull, M. P. Zalucki, A. R. Ives, A. Hulthen, and N. A. Schellhorn. 2017. Estimating the landscape distribution of eggs by Helicoverpa spp., with implications for Bt resistance management. Ecol. Model . 365: 129– 140. Google Scholar CrossRef Search ADS   Pfannenstiel, R. S., and K. V. Yeargan. 1998. Association of predaceous Hemiptera with selected crops. Environ. Entomol . 27: 232– 239. Google Scholar CrossRef Search ADS   R Core Team. 2016. R: A language and environment for statistical computing . R Foundation for Statistical Computing, Vienna, Austria. Reay-Jones, F. P. 2010. Spatial and temporal patterns of stink bugs (Hemiptera: Pentatomidae) in wheat. Environ. Entomol . 39: 944– 955. Google Scholar CrossRef Search ADS PubMed  Reisig, D. D., M. Roe, and A. Dhammi. 2013. Dispersal pattern and dispersion of adult and nymph stink bugs (Hemiptera: Pentatomidae) in wheat and corn. Environ. Entomol . 42: 1184– 1192. Google Scholar CrossRef Search ADS PubMed  Rusch, A., R. Chaplin-Kramer, M. M. Gardiner, V. Hawro, J. Holland, D. Landis, C. Thies, T. Tscharntke, W. W. Weisser, C. Winqvist, et al.   2016. Agriculture landscape simplification reduces natural pest control: a quantitative synthesis. Agr. Ecosyst. Environ . 21: 198– 204. Google Scholar CrossRef Search ADS   SAS Institute Inc. 1998. SAS version 6. SAS/STAT User Guide Release, Cary, NC. Schweiger, O., J. P. Maelfait, W. van Wingerden, F. Hendrickx, R. Billeter, M. I. Speelmans, I. Augenstein, B. Aukema, S. Aviron, D. Bailey, et al.   2005. Quantifying the impact of environmental factors on arthropod communities in agricultural landscapes across organizational levels and spatial scales. J. Appl. Ecol . 42: 1129– 1139. Google Scholar CrossRef Search ADS   Sivakoff, F. S., J. A. Rosenheim, P. Dutilleul, and Y. Carrière. 2013. Influence of the surrounding landscape on crop colonization by a polyphagous insect pest. Entomol. Exp. Appl . 49: 11– 21. Google Scholar CrossRef Search ADS   Soria, M. F., P. E. Degrande, A. R. Panizzi, and M. D. Toews. 2017. Economic injury level of the Neotropical brown stink bug euschistus heros (F.) on cotton plants. Neotrop. Entomol . 46: 324– 335. Google Scholar CrossRef Search ADS PubMed  Southwood, T. R. E. 1978. The construction, description, and analysis of age-specific life-tables, pp. 356– 374. In T. R. E. Southwood (ed.), Ecological methods, with particular reference to the study of insect populations , 2nd ed. Chapman and Hall, London, United Kingdom. Taylor, J., and R. J. Tibshirani. 2015. Statistical learning and selective inference. Proc. Natl. Acad. Sci. U. S. A . 112: 7629– 7634. Google Scholar CrossRef Search ADS PubMed  Thies, C., I. Steffan-Dewenter, and T. Tscharntke. 2003. Effects of landscape context and parasitism at different spatial scales. Oikos . 101: 18– 25. Google Scholar CrossRef Search ADS   Tibshirani, R. J., J. Taylor, R. Lockhart, and R. Tibshirani. 2016. Exact post-selection inference for sequential regression procedures. J. Am. Stat. Assoc . 111: 600– 620. Google Scholar CrossRef Search ADS   Tibshirani, R. J., R. Tibshirani, J. Taylor, J. Loftus, and S. Reid. 2017. selectiveInference: Tools for Post-Selection Inference. R package version 1.2.2. https://CRAN. R-project.org/package=selectiveInference Tillman, P. G., and T. E. Cottrell. 2016. Density and egg parasitism of stink bugs (Hemiptera: Pentatomidae) in elderberry and dispersal into crops. J. Insect Sci . 16: 1– 14. Google Scholar CrossRef Search ADS PubMed  Tillman, P. G., T. D. Northfield, R. F. Mizell, and T. C. Riddle. 2009. Spatiotemporal patterns and dispersal of stink bugs (Heteroptera: Pentatomidae) in peanut-cotton farmscapes. Environ. Entomol . 38: 1038– 1052. Google Scholar CrossRef Search ADS PubMed  Tillman, P. G., T. E. Cottrell, R. F. Mizell, and E. Kramer. 2014. Effect of field edges on dispersal and distribution of colonizing stink bugs across farmscapes of the southeast USA. Bull. Entomol. Res . 104: 56– 64. Google Scholar CrossRef Search ADS PubMed  Todd, J. W. 1989. Ecology and behavior of Nezara viridula. Annu. Rev. Entomol . 34: 273– 292. Google Scholar CrossRef Search ADS   Todd, J. W., and D. C. Herzog. 1980. Sampling phytophagous Pentatomidae on soybean, pp. 438– 478. In M. Kogan and D. C. Herzog (eds.), Sampling methods in soybean entomology . Springer-Verlag, New York, NY. Google Scholar CrossRef Search ADS   Tscharntke, T., and R. Brandl. 2004. Plant-insect interactions in fragmented landscapes. Annu. Rev. Entomol . 49: 405– 430. Google Scholar CrossRef Search ADS PubMed  Tscharntke, T., R. Bommarco, Y. Clough, T. O. Crist, D. Kleijn, T. A. Rand, J. M. Tylianakis, S. Nouhuys, and S. Vidal. 2007. Conservation biological control and enemy diversity on a landscape scale. Biol. Control . 43: 294– 309. Google Scholar CrossRef Search ADS   Tscharntke, T., J. M. Tylianakis, T. A. Rand, R. K. Didham, L. Fahrig, P. Batáry, J. Bengtsson, Y. Clough, T. O. Crist, C. F. Dormann, et al.   2012. Landscape moderation of biodiversity patterns and processes - eight hypotheses. Biol. Rev. Camb. Philos. Soc . 87: 661– 685. Google Scholar CrossRef Search ADS PubMed  Turnipseed, S. G., M. J. Sullivan, J. E. Mann, and M. E. Roof. 1995. Secondary pests in transgenic Bt cotton in South Carolina, pp 768– 769. 1995 Proceedings Beltwide Cotton Conferences, San Antonio, TX, USA, January 4-7, Vol 2. Memphis: National Cotton Council. Werling, B. P., T. D. Meehan, B. A. Robertson, C. Gratton, and D. A. Landis. 2011. Biocontrol potential varies with changes in biofuel-crop plant communities and landscape perenniality. GCB Bioenergy . 3: 347– 359. Google Scholar CrossRef Search ADS   Wilson, B. P., A. J. Morris, B. E. Arroyo, S. C. Clark, and R. B. Bradbury. 1999. A review of the abundance and diversity of invertebrate and plant foods of granivorous birds in northern Europe in relation to agricultural change. Agr. Ecosyst. Environ . 65: 13– 30. Google Scholar CrossRef Search ADS   Yasuda, M., T. Mitsunaga, A. Takeda, K. Tabuchi, K. Oku, T. Yasuda, and T. Watanabe. 2011. Comparison of the effects of landscape composition on two mirid species in Japanese rice paddies. Appl. Entomol. Zool . 46: 519– 525. Google Scholar CrossRef Search ADS   Zeilinger, A. R., D. M. Olson, and D. A. Andow. 2011. Competition between stink bug and heliothine caterpillar pests on cotton at within-plant spatial scales. Entomol. Exp. Appl . 141: 59– 70. Google Scholar CrossRef Search ADS   Published by Oxford University Press on behalf of Entomological Society of America 2018. This work is written by (a) US Government employee(s) and is in the public domain in the US.

Journal

Environmental EntomologyOxford University Press

Published: Apr 4, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off