Ecoinformatics Can Reveal Yield Gaps Associated with Crop-Pest Interactions: A Proof-of-Concept

Jay A. Rosenheim; Matthew H. Meisner

doi:10.1371/journal.pone.0080518

Abstract

Farmers and private consultants execute a vast, decentralized data collection effort with each cropping cycle, as they gather pest density data to make real-time pest management decisions. Here we present a proof of concept for an ecoinformatics approach to pest management research, which attempts to harness these data to answer questions about pest-crop interactions. The impact of herbivory by Lygus hesperus on cotton is explored as a case study. Consultant-derived data satisfied a ‘positive control’ test for data quality by clearly resolving the expected negative relationship between L. hesperus density and retention of flower buds. The enhanced statistical power afforded by the large ecoinformatics dataset revealed an early-season window of crop sensitivity, during which L. hesperus densities as low as 1-2 per sample were associated with yield loss. In contrast, during the mid-season insecticide use by farmers was often unnecessary, as cotton compensated fully for moderate L. hesperus densities. Because the dataset emerged from the commercial production setting, it also revealed the limited degree to which farmers were willing to delay crop harvest to provide opportunities for compensatory fruiting. Observational approaches to pest management research have strengths and weaknesses that complement those of traditional, experimental approaches; combining these methods can contribute to enhanced agricultural productivity.

Citation: Rosenheim JA, Meisner MH (2013) Ecoinformatics Can Reveal Yield Gaps Associated with Crop-Pest Interactions: A Proof-of-Concept. PLoS ONE 8(11): e80518. https://doi.org/10.1371/journal.pone.0080518

Editor: Raul Narciso Carvalho Guedes, Federal University of Viçosa, Brazil

Received: July 8, 2013; Accepted: October 3, 2013; Published: November 15, 2013

Copyright: © 2013 Rosenheim, Meisner. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Funding: This project was supported in part by grants from the California State Support Committee of Cotton Incorporated; the University of California Statewide IPM Program; USDA-NRICGP grant 2006-01761; and a contract from the California Department of Pesticide Regulation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: Part of the financial support for this research came from a commercial source, the California State Support Committee of Cotton Incorporated. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.

Introduction

A growing world population will necessitate a substantial increase in global agricultural production [1]. Part of this increase can come from more effective suppression of damaging pest populations [2]. At the same time, indiscriminate use of pesticides is undesirable, because of negative consequences for human and environmental health, and because of the potential to accelerate the evolution of resistance and disrupt ecosystem services contributed by natural enemies [3]. Thus, we need an efficient means of distinguishing between conditions under which pests cause yield shortfalls versus conditions under which crop plants can compensate for herbivory, rendering pesticide applications or other pest control interventions unnecessary.

Researchers have traditionally relied on manipulative experiments to identify conditions under which pests depress yield. Although experimentation will continue to be crucial, alternatives to experimentation that have hitherto received little attention may play a complementary role. In particular, observational approaches may be useful in providing greater statistical power for detecting the small, but economically important, yield losses that drive farmer decision-making [4]. Key yield losses are often small in magnitude for two reasons. First, most crops are impacted by many pests, including diverse arthropod, nematode, pathogen, and weed populations, each of which may generate small effects, but which can, collectively, depress yield substantially. Because pests are generally managed individually or in small groups rather than in aggregate, we need to resolve individual effects. Second, to minimize unnecessary pesticide applications, researchers must be able to resolve the yield loss decision point at which a profit maximizing farmer will opt to apply a pesticide to protect crop yield (the “economic injury level”; [5]). Because the cost of a pesticide application is often small relative to the value of the crop, this decision point is often reached when only a tiny loss of yield (ca. 2%) is threatened; experiments generally do not successfully resolve effects of this size [4]. In some cases, researchers have formed consortia to replicate experiments extensively in space and time, creating composite data sets that enhance statistical power. This approach has had notable success [6-8], but can only be used when substantial research resources are available, and as such can be used only for the most important pests in developed countries and perhaps only very infrequently in developing countries.

Ecoinformatics, which involves mining pre-existing data sources to generate large, observational data sets, holds promise as a lower-cost means of generating higher-resolution insights into pest impacts on crop yield. Farmers monitor pest densities in their crops to guide real-time pest management decision-making. This vast, decentralized data collection effort can be combined with farmer data on realized crop yields to generate large observational datasets quantifying crop-pest interactions.

Here we present a proof of concept for the application of ecoinformatics to the study of pest impacts on crop yield. A long-standing controversy surrounds the possible impact of Lygus hesperus Knight (Heteroptera: Miridae) on cotton production in the Western United States. L. hesperus prefer to feed on young flower buds on cotton, and plants may respond by abscising damaged buds [9-11]). However, decades of experimentation, focused nearly exclusively on mid-season (July) L. hesperus herbivory, have resolved yield effects only at high pest densities (>8 insects per standard sample of 50 sweeps; [12-17]). Cotton plants exhibit plasticity in growth; thus, plants can potentially compensate for herbivory that causes early flower bud abscission by increasing the subsequent production of ‘replacement’ flower buds, although this may delay fruit maturation and crop harvest [15,18,19]. Farmers, however, remain skeptical of university recommendations for high L. hesperus treatment thresholds. Farmers may be unwilling to extend the cotton growing season into the fall, when a variety of crop risks may be present, and generally apply insecticides at much lower densities (2-4 L. hesperus per sample). Furthermore, farmers generally use a sliding threshold: earlier in the growing season (pre-flowering; roughly June in California), farmers often use a threshold near 2 bugs per sample, whereas later in the growing season (roughly July), farmers often use a threshold near 4 bugs per sample. Farmers rarely suppress L. hesperus after early August, when flower buds have diminished chances of developing to mature fruit prior to harvest.

A renewed focus on L. hesperus is timely for two reasons. First, a new cotton species, Gossypium barbadense L. (“Pima cotton”), has replaced Gossypium hirsutum L. (“upland cotton”) as the dominant species cultivated in California, and it is not known if the impacts of L. hesperus on these two cotton species are similar. Second, the widespread adoption of transgenic cotton cultivars expressing endotoxin genes from Bacillus thuringiensis and the associated withdrawal of broad-spectrum insecticides targeting lepidopteran pests has resulted in the emergence of Lygus spp. and other hemipteran pests as primary threats in many of the world’s cotton producing regions [8,20,21].

To build a proof of concept, we will first address a primary perceived obstacle to this approach, namely that farmer- or consultant-derived data would be too heterogeneous or of insufficient basic quality to reveal signals of herbivore impact on crop performance. Although L. hesperus impact on cotton yield is controversial, there is no question that L. hesperus contributes to flower bud abscission [9-11,15,22,23]. We will therefore ask if a farmer-derived dataset can resolve this expected effect as a positive control for data quality. We will then use the farmer-derived data set to evaluate the possibility that crop sensitivity varies across the growing season by exploring associations between early- (June) and mid-season (July) herbivory by L. hesperus and (i) cotton yield and (ii) the time when farmers opt to terminate crop growth in preparation for harvest.

Materials and Methods

The data set was built exclusively by collecting pre-existing data derived from observations of commercial cotton production in California’s San Joaquin Valley. Data streams included the following:

Lygus hesperus densities. L. hesperus population densities were sampled by four independent pest control consulting firms, who are employed by farmers to provide pest monitoring services and control recommendations. The final data set included observations for 1432 cotton crops produced by 35 farms between 1997 and 2008. All consultants used the same standard sampling procedure: an approximately weekly series (typically 6-12) of sweep net samples, each comprising 50 swings of a sweep net across the top of the plant canopy. Successive samples were transformed into mean L. hesperus density estimates by calculating the area under the curve of L. hesperus density versus time, using linear interpolation between successive density estimates and dividing by the total duration of the sampling interval; this accounted for the sometimes uneven time intervals between successive samples. Counts reflect all motile stages combined, as not all consultants reported separate counts of nymphs and adults.
Cotton yield. Farmers shared data on cotton lint yield (kg/ha) for years for which L. hesperus density estimates were available. Yield data were available for 1118 cotton crops.
Historical cotton yield. Field to field variation in yield potential across the cotton growing region of the San Joaquin Valley (from Madera and Merced Counties in the north to Kern County in the south) can be substantial, due to differences in soil quality and local climate. Stable differences in farmer growing practices also contribute to between-field variation in yield. To control statistically for variable yield potential, we requested from farmers records on historical cotton yields for each of their fields as far back as their records permitted, or until 1990. With these historical data, the following quantities were calculated:
$Expected yield deviation = \sum_{i \neq f}^{} \frac{y_{i} - {\bar{y}}_{i}}{N},$

where y_i is the yield observed in a particular field during year i, ${\bar{y}}_{i}$ , is the mean yield observed for year i across the entire San Joaquin Valley, as reported by the USDA (http://www.nass.usda.gov), and N is the number of years between 1990 and 2008, other than the focal year, year f, for which yield estimates were available for the cotton species in question;

Observed yield deviation = y_{f} - {\bar{y}}_{f},

where y_f is the yield observed for a particular field-year for which L. hesperus density estimates were available and ${\bar{y}}_{f}$ is the mean yield observed for that year across the entire San Joaquin Valley; and finally:

Yield gap = Observed yield deviation - Expected yield return .

The yield gap was the response variable for analyses of yield effects.

Flower bud retention. Pest management consultants and, in some cases, independent agronomy consultants provided estimates of flower bud retention, produced by scoring retention at the first position on fruiting branches located on the top five mainstem nodes of a sample of cotton plants.
Pesticide applications. The dates and pest targets of all pesticide applications were obtained from consultants and the California Department of Pesticide Regulation’s on-line Pesticide Use Reporting system (http://www.cdpr.ca.gov/docs/pur/purmain.htm). Pesticide applications made at planting (i.e., aldicarb) provided data on planting date.
Plant growth regulator applications. The plant growth regulator mepiquat chloride is used by farmers to shift the plant’s resource allocations from vegetative to reproductive growth, which may enhance yield [24]. Because L. hesperus herbivory can trigger increased use of mepiquat chloride [25], we obtained data on mepiquat chloride from consultants.
Date of first application of chemical defoliant. Cotton must be defoliated prior to harvest. The date of first defoliant application was used as an indicator of the time when the farmer opted to terminate the crop. Defoliant application dates were obtained from pest control consultants and the California Department of Pesticide Regulation. To combine data across years, dates were expressed as [(date of first defoliant application) – (mean date of first defoliant application for the year in question)]; thus, positive numbers indicate application dates that were later than average. These data were used to quantify farmer willingness to extend the growing season in response to L. hesperus damage to allow for plant compensation.
Weather. Temperature and precipitation have the potential to affect both cotton yield and L. hesperus populations. We obtained the mean monthly temperature and precipitation from weather stations in the San Joaquin Valley from the National Oceanic and Atmospheric Administration's National Climatic Data Center (http://www.ncdc.noaa.gov). For each record in the database, we used climatic data from the nearest weather station (mean ± SD distance to station, 13.1 ± 7.2 km) with available data for that year.

Data were managed in a relational database programmed in SQL Server and accessed using an interface designed by a private software developer (“Cottonformatics”, Ten2Eleven Business Solutions).

Statistical analysis

Both the empirical record and theoretical treatments of plant-animal interactions suggest that the relationship between herbivore densities and host plant performance will often be non-linear, with losses accelerating as plant damage increases [26-28]. L. hesperus impacts on cotton were therefore analyzed using a flexible non-linear regression method, generalized additive models (GAM), implemented using program mgcv 1.7-6 in R [29]. GAMs are attractive because they allow the data “to speak for themselves” in suggesting the form of the function, rather than imposing a particular model (linear, quadratic, etc.) on the data. GAMs also provide an objective means of avoiding the over-fitting of the data (i.e., fitting the noise instead of the underlying trends) through a process that penalizes excessive “wiggliness” of the resulting function, as quantified by the second derivative of the curve (generalized cross validation). To keep the analyses as transparent as possible, we opted to analyze quite simple models; three response variables (flower bud retention, yield gap, and the date of the first application of defoliant) were analyzed separately, and in all cases the models included main effects of farm, year, Gossypium species, and mean L. hesperus densities during specified time intervals. To model the date of the first defoliant application, we also included mean local monthly temperature and precipitation variables. The overall modeling approach was semi-parametric: only the L. hesperus density variables were smoothed, using non-parametric thin plate regression splines, with the remaining variables treated as traditional linear additive effects in the usual parametric fashion. All analyses employed a Gaussian distribution and the identity link function.

To explore the possibility that other factors, correlated with both L. hesperus densities and cotton yield, might create spurious associations between L. hesperus and cotton performance, we explored GAM models including other variables. Because not all variables were measured for all records, we could not build a single model with all factors considered simultaneously without sacrificing many data records. We therefore conducted a series of analyses to explore different sets of possible confounders. Although it is not possible to exclude the possibility of unmeasured confounding variables, we attempted to search diligently for possible hidden sources of causality. Because insecticides applied to control L. hesperus may trigger secondary pest outbreaks that could depress yield [30], we explored a model including the season-long number of insecticide applications that targeted L. hesperus. Because wet years are thought to be associated with both higher L. hesperus densities [31] and delayed planting, causing lower yields, we explored a model that included planting date, expressed as a deviation from the mean planting date for the year in question. We also explored a model that included mean local monthly temperature and precipitation values. Because L. hesperus densities may be lower in larger fields [32,33], and because larger fields may be correlated with agricultural intensification, and thus be associated with higher yields, we explored a model that included field size (ha). Because L. hesperus densities might be correlated with densities of other herbivores, including spider mites (Tetranychus spp.), aphids (Aphis gossypii Glover), whiteflies (Bemisia tabaci [Gennadius] and Trialeurodes vaporariorum [Westwood]), thrips (Frankliniella occidentalis [Pergande]), and Lepidopteran caterpillars (various, but primarily Spodoptera spp.), we explored a model that included the numbers of pesticide applications targeting each of these herbivores. Finally, because L. hesperus herbivory may change plant growth form, eliciting changes in farmer decisions to apply plant growth regulators (mepiquat chloride) or defoliants, each of which may influence yield [25], we explored a model including the number of these applications.

An additional concern is that our GAM analyses might not isolate correctly the associations between June versus July L. hesperus populations and cotton yield if insect densities are strongly correlated in time. June and July L. hesperus densities were positively correlated across our full data set, but the association was not tight, perhaps in part due to insecticide use (Pearson’s correlation, R² = 0.29, N = 1432, P < 0.0001). To determine if this modest correlation might be generating interpretational errors, we performed stepwise multiple linear regression analyses examining the effects of June, July, and June x July L. hesperus densities on yield. By comparing different orders of variable inclusion, we evaluated the possibility that L. hesperus densities during June might mask important correlates of later herbivory. This analysis also allowed us to test for an interaction between June and July L. hesperus (GAMs assume additivity).

Results

A positive control for data quality

The quality of the farmer-derived data was sufficiently high that the expected negative impact of L. hesperus on flower bud retention was clearly revealed (Figure 1). During June, the full GAM model explained 59.5% of the deviance (Table 1), roughly half of which was attributable to L. hesperus (a model including only L. hesperus densities explained 30.5% of the deviance). In July the full model similarly explained 66% of the deviance (Table 2), but only a quarter of this was now attributable to L. hesperus (a model including only L. hesperus explained 16.4% of the deviance). Thus, the ecoinformatics data set passed this initial “positive control” test. The flower bud retention data furthermore suggested that the underlying damage generated by L. hesperus was similar in June and July; in both cases, flower bud retention dropped by ca. 10% as L. hesperus densities increased to ca. 4 insects/sample.

Download:

Figure 1. Flower bud retention and L. hesperus density.

Association between the proportion of cotton flower buds retained and mean L. hesperus densities observed during (A) June, and (B) July. Retention values are residuals after controlling statistically for effects of farm, year, and cotton species. The solid lines are the smooths from the GAM models, and the shaded regions are the 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0080518.g001

Term	df	F	P
Farm	25	7.56	<1x10^-15
Year	11	15.00	<1x10^-15
Gossypium species	2	1.48	0.23
June L. hesperus density	2.37	27.45	<1x10^-15

Table 1. Generalized additive model of factors associated with early-season (June) retention of flower buds by cotton.

Deviance explained = 59.5%, N = 656

CSV

Download CSV

Term	df	F	P
Farm	24	9.61	<1x10^-15
Year	11	27.02	<1x10^-15
Gossypium species	2	5.76	0.003
July L. hesperus density	4.21	13.3	1.1x10^-12

Table 2. Generalized additive model of factors associated with mid-season (July) retention of flower buds by cotton.

Deviance explained = 66.0%, N = 561

CSV

Download CSV

Lygus hesperus densities and Gossypium yield

The ecoinformatics data set revealed an association between early-season (June) L. hesperus herbivory and depressed cotton yield that is expressed even at very low herbivore densities (1-2 L. hesperus per sweep sample; Table 3, Figure 2). This result did not rely on the highest L. hesperus density observations; repeating the analysis excluding the records (n = 88) with mean June L. hesperus densities >2.0 revealed the same significant yield decline associated with June herbivory (df = 2.6, F = 6.13, P = 0.0002). In contrast, the data revealed no evidence for any association between L. hesperus and yield later during the fruiting period (July), across the full range of L. hesperus densities represented in the data set (0-10 L. hesperus per sample). Repeating the analysis with the growing season broken into successive two-week periods suggested that cotton’s sensitive period may extend very weakly into the first half of July; thereafter, a non-significant trend towards increasing yield with increasing L. hesperus densities was observed (Table S1, Figure S1). Farmers delayed the first application of defoliant in fields that harbored moderate to heavy L. hesperus populations during 1-15 July (P = 0.00033; note that this effect may have reversed for the handful of fields that had the very highest L. hesperus densities). However, no such delays in crop defoliation were found to be associated with L. hesperus densities during June or 16-31 July (Figure 3, Table S2). Thus, during July, but not June, the plant appears to compensate strongly for the abscission of flower buds, perhaps aided by modest extensions of the crop’s growing period in response to early July herbivory.

Term	df	F	P
Farm	35	2.44	8.3x10^-6
Year	10	12.56	<1x10^-15
Gossypium species	1	0.10	0.76
June L. hesperus density	6.39	9.24	7.7x10^-12
July L. hesperus density	2.85	0.77	0.53

Table 3. Generalized additive model of factors associated with yield of cotton, Gossypium spp.

Deviance explained = 22.0%, N = 1118

CSV

Download CSV

Download:

Figure 2. Cotton yield gap and L. hesperus density.

Association between the cotton yield gap (difference between the observed and expected yield, after controlling for annual fluctuations in valley-wide yield) and mean L. hesperus densities observed during (A) June, and (B) July. The solid lines are the smooths from the GAM models, and the shaded regions are the 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0080518.g002

Download:

Figure 3. Crop termination and L. hesperus density.

Association between the date of the first application of a defoliant to the cotton crop (expressed as a deviation from the year’s mean date of initial defoliant treatment) and mean L. hesperus densities observed during (A) 1-15 June, (B) 16-30 June, (C) 1-15 July, and (D) 16-31 July. The solid lines are the smooths from the GAM models, and the shaded regions are the 95% confidence intervals. Positive values for the date of defoliation indicate dates later than the mean for that year. Note that because the GAM penalizes the second derivative of the fitted curve, the model collapses to linearity when evidence for non-linearity is sufficiently weak, as is seen in panel (B). See Table S2 for the GAM summary.

https://doi.org/10.1371/journal.pone.0080518.g003

As expected given the long list of factors that generate variation in cotton yield, densities of L. hesperus explained only a very small portion of the total observed yield variation (a model including only L. hesperus densities in June and July explained 7.8% of total deviance). Nevertheless, the large sample size (N = 1118 records) allowed this economically important effect to be resolved readily.

Collinearity of June and July L. hesperus densities

Although our GAM analyses did support some non-linearities in the relationships between June L. hesperus and cotton yield, the overall trends were broadly linear. We therefore used stepwise multiple linear regression to examine if correlations between June and July L. hesperus densities might be masking effects of July herbivory. We forced Farm, Year, and Gossypium species into the model first. Using a threshold P-value to enter = 0.25 and a threshold P-value to remove = 0.10, June L. hesperus density was selected as the first and only variable to be added (P < 0.0001, R² = 0.065, ΔAICc = -73.4). July and June x July L. hesperus densities were not entered, despite the permissive threshold. If instead we forced July L. hesperus into the model as the first step, we obtained a modest improvement in fit (P = 0.0002, R² = 0.013, ΔAICc = -12.2), but the automated stepwise procedure then added June L. hesperus (P < 0.0001, R² = 0.066, ΔAICc = -60.6), and as the final step removed July L. hesperus (P = 0.24, R² = 0.065, ΔAICc = -0.6). Thus, we found no evidence that the modest correlation between June and July L. hesperus densities was hiding an underlying effect of July L. hesperus herbivory on crop yield.

Searching for possible confounding variables

Statistical models including variables we deemed most viable as candidate confounders did not change the underlying relationship between L. hesperus densities and cotton yield. GAM models including the number of insecticide applications targeting L. hesperus (Table S3), the field’s planting date (Table S4), the field’s size (Table S5), the number of insecticide applications targeting other herbivores (Table S6), the number of applications of plant growth regulators or defoliants (Table S7), and mean local monthly temperature and precipitation (Table S8) all supported the same core inference: June L. hesperus were consistently associated with depressed cotton yield, whereas July L. hesperus were not.

Discussion

The goal of this study was to develop a proof of concept for the application of ecoinformatics methods to quantifying pest impacts on crop yield, focusing on the impact of L. hesperus on cotton. L. hesperus was chosen as a test case, because, despite decades of experimentation, its optimal management remains controversial. Furthermore, the prospect of further exclusive reliance on experimentation was daunting: L. hesperus is so mobile that very large plots are required to maintain insect density manipulations, and secondary pest outbreaks are often associated with use of insecticides to manipulate L. hesperus densities, making interpretation of treatment effects difficult (e.g., [12,13]). The farmer- and consultant-derived data set passed an initial “positive control” test by resolving the expected impact of L. hesperus on flower bud retention (Figure 1). L. hesperus explained 30.5% (June) and 16.4% (July) of the deviance in flower bud retention. These values are comparable to three earlier researcher-generated data sets that examined the correlation between Lygus spp. densities and flower bud abscission across commercial upland cotton fields (r² = 0.21 [7]; r² = 0.22-0.44 [9] Leigh, Kerby & Wynholds 1988]; and r² = 0.19 [data reanalyzed from [34]]). Although data quality and heterogeneity will always be important concerns in ecoinformatics studies, these results encourage the hope that consultant-derived data can support useful research inferences.

Yield gaps: evidence from modeling and experimentation

Our analyses sought to identify conditions under which L. hesperus might be associated with unrecognized yield losses. Indeed, farmers appear to be losing yield when cotton harbors even very low L. hesperus densities during June, when cotton plants are producing their first flower buds.

Our confidence that the association between June L. hesperus populations and loss of yield reflects a direct causal relationship would be bolstered by complementary evidence from other sources. We adduce two such sources of supporting evidence. First, two simulation models of Gossypium growth and fruiting, each parameterized with observations from commercial cotton farming in California’s San Joaquin Valley, but otherwise quite different in structure, each predicted that Gossypium would be most sensitive to flower bud loss early during the reproductive period ([22,35]). Thus, there was a clear reason, a priori, to expect that June L. hesperus herbivory would depress Gossypium yield.

Second, a careful reexamination of the published experimental literature reveals suggestive evidence that early (June) L. hesperus herbivory depresses cotton yield. To our knowledge, only once have researchers manipulated June L. hesperus densities without continuing the manipulations later in the growing season. A June-only manipulation of L. hesperus densities is important, because season-long suppression of L. hesperus with insecticides often triggers secondary pest outbreaks, making such experiments hard to interpret (e.g., [12,13]). Falcon et al. [13] in their ‘1969 Experiment’ implemented a low L. hesperus density treatment (“5 early”) by applying insecticides any time L. hesperus densities reached 5 per sweep sample during the earliest period of fruiting (<5 blooms/4 m of plants; roughly June, in some plots extending until July 8). The manipulation was then relaxed for the remainder of the growing season. Mean June L. hesperus density per sweep sample in the ‘5 early’ treatment plots was 1.48, substantially less than the 5.50 observed in the untreated control. To achieve these clear density differences during June, when nearly all L. hesperus are present as mobile, winged adults (86% adults in this experiment), required the use of very large plots: plot dimensions were 201 x 805 m. As a result, the level of replication was small (n = 4), and statistical power was limited. Nevertheless, the first cotton harvest revealed a statistically significant 10.7% yield loss in the untreated control compared to plots where L. hesperus densities were suppressed (P < 0.05). The second harvest was very small in comparison to the first harvest (producing just 6.1% more yield), and changed the picture only slightly: early L. hesperus were still associated with a 9.3% loss of yield. Nevertheless, inclusion of the second harvest resulted in the loss of the statistical significance of the yield gap (P > 0.05). The original authors discussed these highly suggestive trends, and called for further investigations of the yield gaps stemming from early L. hesperus herbivory.

Thus, three forms of evidence, simulation modeling, manipulative experimentation, and the observational study that we report here, support the same inference. Cotton plants appear to have a window of heightened sensitivity to L. hesperus herbivory expressed early during the fruiting period.

Unnecessary use of insecticides

Our analyses also identified conditions under which farmers appear to be applying insecticides unnecessarily to suppress economically benign L. hesperus populations. L. hesperus populations after mid-July do not appear to be associated with a detectable loss of yield. The mean density of L. hesperus that triggered the application of insecticides during late July was 4-5 bugs per sweep sample (Figure 4), comfortably within the range of densities that showed no effect on cotton yield (Figure 2B). Now that the early window of crop susceptibility has been identified, the hope is that farmers will become more receptive to the message that, according to their own data, cotton can compensate for moderate densities of L. hesperus later, after the sensitive period closes.

Download:

Figure 4. Insecticide use decisions and L. hesperus density.

Mean ± SE density of L. hesperus that triggered the application of insecticides targeting L. hesperus (solid circles) during successive 2-week intervals during cotton’s fruiting period. Also shown for comparison are the mean L. hesperus densities observed across all fields (open circles). Numbers above the symbols are the number of fields for which density estimates were available.

https://doi.org/10.1371/journal.pone.0080518.g004

Why might the sensitivity of cotton to L. hesperus decrease as the plant grows? The fruit maturation process in cotton appears to shift from being sink-limited to source-limited as the plant grows [15,18,23,35,36]. Young cotton plants have well-developed resource gathering structures (leaves and roots), but few flower buds. In this case, plants protected from herbivory or other stresses can retain and mature nearly 100% of flower buds initiated, and any flower bud abscission in response to L. hesperus feeding may translate into an immediate loss of yield (Figure 2A). In contrast, maturing plants rapidly produce a large number of flower buds, whose demands increasingly outstrip the plant’s resource supply. As a result, even in the absence of any environmental stress, abscission of flower buds or young fruit rapidly increases beginning around the time of flowering [18,23,35-37]. In this case, we may expect little effect of L. hesperus on yield as long as the proportion of flower buds damaged does not exceed that which the plant would abscise anyway [18]. Thus, the relationship between L. hesperus densities during July and cotton yield may be non-linear: an initial range of L. hesperus densities that generate no effect on yield is expected, as was found (Figure 2B). A subsequent phase of yield decline is, however, also expected as L. hesperus damages a proportion of flower buds that is greater than the proportion that would have been abscised in the absence of damage. Experiments suggest that this threshold is reached at densities of ca. 10 L. hesperus per sweep sample.

This highlights an important limitation of the ecoinformatics approach: because farmers manage L. hesperus with insecticides in a universally aggressive manner, L. hesperus densities >10 per sample were never observed, even in a sample of >1000 commercial cotton fields (Figure 2). Thus, these farmer-derived data cannot be used to refine the critical late July L. hesperus density threshold at which yield loss begins. Observational research approaches like the one described here can only work with pre-existing variation. Thus, observational studies cannot explore scenarios like very high L. hesperus densities if, as was observed here, farmers exclude them the commercial setting. Experimentation has an obvious and important advantage in this regard.

A model of cotton yield that includes not only L. hesperus densities but also the number of insecticide applications targeting L. hesperus suggests that some additional yield loss (29.6 ± 10.8 kg/ha) is associated with each insecticide application, above and beyond the effects of L. hesperus herbivory (Table S3). This additional yield loss may reflect the disruptive effects of insecticide use on natural enemies, which can precipitate damaging outbreaks of other pests [30]. Thus, some July insecticide applications targeting L. hesperus may be not only unnecessary and costly, but even counterproductive.

Plant compensation by extending the growing season

The analysis presented here suggests that cotton is sensitive to very low densities of L. hesperus pre-flowering. Why couldn’t cotton compensate by developing additional fruit later in the growing season? Compensation ability is often treated as a plant trait, but the termination of crop growth is partially a response of the crop plant and partially a decision made by the farmer. Farmers have potent disincentives for delaying crop harvest, including the risk of fall rain, which degrades lint quality and can prevent harvest; the risk that late-season aphid or whitefly populations will build, producing honeydew that contaminates lint and reduces crop marketability; the cost of extra irrigation water needed to extend crop growth; and the pressing need to prepare fields for the following crop before rains arrive. For these reasons, an ecoinformatics approach that captures not just plasticity in plant growth, but also flexible farmer decision making is essential to measuring actual opportunities for plant compensation through delayed crop termination. Farmers delayed the termination of crops that harbored high L. hesperus populations during early July, but no such association was observed with June L. hesperus populations, perhaps contributing to the early window of crop sensitivity (Figure 3).

Conclusions

Decentralized data collection in the agricultural sector can generate large datasets that hold the promise of greater statistical power. A proof of concept for an ecoinformatics approach focused on L. hesperus interactions with cotton suggested that the quality of consultant-generated data, as assessed by quantifying the relationship between L. hesperus and flower bud retention, was comparable to that of researcher-generated data. Strengths of the approach included (i) the ability to resolve small, but economically important yield gaps associated with an early window of crop sensitivity, and (ii) an opportunity to capture both plant and farmer responses to L. hesperus herbivory. A key weakness of the approach was the inability to consider the effects of very high L. hesperus densities, as these are excluded from the commercial farming setting by uniformly aggressive farmer pest suppression. The use of ecoinformatics in several subdisciplines of the agricultural sciences is growing [38-43], providing a research methodology that can complement traditional experimental approaches to enhancing agricultural productivity.

Supporting Information

Figure S1.

Association between the cotton yield gap (difference between the observed and expected yield, after controlling for annual fluctuations in valley-wide yield) and mean L. hesperus densities observed during (A) 1-15 June, (B) 16-30 June, (C) 1-15 July, and (D) 16-31 July. The solid lines are the smooths from the GAM models, and the shaded regions are the 95% confidence intervals. Note that because the GAM penalizes the second derivative of the fitted curve, the model collapses to linearity when evidence for non-linearity is sufficiently weak, as is seen in panels (A), (B), and (D).

https://doi.org/10.1371/journal.pone.0080518.s001

(EPS)

Table S1.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., with the fruiting season broken into successive 2-week intervals.

https://doi.org/10.1371/journal.pone.0080518.s002

(DOCX)

Table S2.

Generalized additive model of factors associated with the date of the first application of defoliant to cotton, Gossypium spp., including both L. hesperus densities and the monthly mean temperatures and rainfall beginning October of the year preceding the spring planting of the focal cotton crop.

https://doi.org/10.1371/journal.pone.0080518.s003

(DOCX)

Table S3.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including both L. hesperus densities and the number of insecticide applications that targeted L. hesperus.

https://doi.org/10.1371/journal.pone.0080518.s004

(DOCX)

Table S4.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including both L. hesperus densities and planting date.

https://doi.org/10.1371/journal.pone.0080518.s005

(DOCX)

Table S5.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including both L. hesperus densities and field size.

https://doi.org/10.1371/journal.pone.0080518.s006

(DOCX)

Table S6.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including L. hesperus densities and the number of insecticide applications made targeting spider mites (Tetranychus spp.), aphids (Aphis gossypii), whiteflies (Bemisia tabaci and Trialeurodes vaporariorum), thrips (mostly Frankliniella occidentalis), and various Lepidoptera (Spodoptera spp. and others).

https://doi.org/10.1371/journal.pone.0080518.s007

(DOCX)

Table S7.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including both L. hesperus densities and the numbers of applications of either plant growth regulators or defoliants.

https://doi.org/10.1371/journal.pone.0080518.s008

(DOCX)

Table S8.

Generalized additive model of factors associated with yield of cotton, Gossypium spp., including both L. hesperus densities and the monthly mean temperatures and rainfall beginning October of the year preceding the spring planting of the focal cotton crop.

https://doi.org/10.1371/journal.pone.0080518.s009

(DOCX)

Acknowledgments

We extend sincere thanks to the community of private consultants and cotton farmers who generously shared their time and data. Lawrence Wilhoit provided help with obtaining data from the California Department of Pesticide Regulation’s pesticide use reporting database. Peter Goodell provided advice during the earliest stages of the project. Sarina Jepsen, Andrew Zink, Crystal Perreira, Mai Nguyen, Lea Bateman, and Lindsey Hack assisted in building the database. Jim Jones, Jason Kuechler, and the staff at Ten2Eleven Business Solutions developed the Cottonformatics database application. Neil Willits consulted on the statistical analyses. Peter Goodell, Lindsey Hack, Vince Jones, Kimberly Steinmann, and Lawrence Wilhoit provided constructive suggestions on an earlier draft of the paper.

Author Contributions

Conceived and designed the experiments: JAR MHM. Performed the experiments: JAR MHM. Analyzed the data: JAR MHM. Contributed reagents/materials/analysis tools: JAR MHM. Wrote the manuscript: JAR MHM.

References

1. Godfray HCJ, Beddington JR, Crute IR, Haddad L, Lawrence D et al. (2010) Food security: the challenge of feeding 9 billion people. Science 327: 812-818. doi:https://doi.org/10.1126/science.1185383. PubMed: 20110467.
- View Article
- Google Scholar
2. Oerke EC (2006) Crop losses to pests. J Agric Sci 144: 31-43. doi:https://doi.org/10.1017/S0021859605005708.
- View Article
- Google Scholar
3. Birch ANE, Begg GS, Squire GR (2011) How agro-ecological research helps to address food security issues under new IPM and pesticide reduction policies for global crop production systems. J Exp Bot 62: 3251-3261. doi:https://doi.org/10.1093/jxb/err064. PubMed: 21669880.
- View Article
- Google Scholar
4. Rosenheim JA, Parsa S, Forbes AA, Krimmel WA, Law YH et al. (2011) Ecoinformatics for integrated pest management: expanding the applied insect ecologist’s tool-kit. J Econ Entomol 104: 331-342. doi:https://doi.org/10.1603/EC10380. PubMed: 21510177.
- View Article
- Google Scholar
5. Stern VM, Smith RF, van den Bosch R, Hagen KS (1959) The integrated control concept. Hilgardia 29: 81-101.
- View Article
- Google Scholar
6. Ragsdale DW, McCornack BP, Venette RC, Potter BD, MacRae IV et al. (2007) Economic threshold for soybean aphid (Hemiptera: Aphididae). J Econ Entomol 100: 1258-1267. doi:https://doi.org/10.1603/0022-0493(2007)100[1258:ETFSAH]2.0.CO;2. PubMed: 17849878.
- View Article
- Google Scholar
7. Musser FR, Lorenz GM, Stewart SD, Bagwell RD, Leonard BR et al. (2009a) Tarnished plant bug (Hemiptera: Miridae) thresholds for cotton before bloom in the midsouth of the United States. J Econ Entomol 102: 2109-2115. doi:https://doi.org/10.1603/029.102.0614. PubMed: 20069839.
- View Article
- Google Scholar
8. Musser FR, Catchot AL, Stewart SD, Bagwell RD, Lorenz GM et al. (2009b) Tarnished plant bug (Hemiptera: Miridae) thresholds and sampling comparisons for flowering cotton in the midsouthern United States. J Econ Entomol 102: 1827-1836. doi:https://doi.org/10.1603/029.102.0513. PubMed: 19886447.
- View Article
- Google Scholar
9. Leigh TF, Kerby TA, Wynholds PF (1988) Cotton square damage by the plant bug, Lygus hesperus (Hemiptera: Heteroptera: Miridae), and abscission rates. J Econ Entomol 81: 1328-1337.
- View Article
- Google Scholar
10. Forbes AA, Rosenheim JA (2011) Plant responses to insect herbivore damage are modulated by phosphorus nutrition. Entomol Exp Appl 139: 242-249. doi:https://doi.org/10.1111/j.1570-7458.2011.01125.x.
- View Article
- Google Scholar
11. Cooper WR, Spurgeon DW (2012) Injury to cotton by adult Lygus hesperus (Hemiptera: Miridae) of different gender and reproductive states. Environ Entomol 41: 342-348. doi:https://doi.org/10.1603/EN11236. PubMed: 22507007.
- View Article
- Google Scholar
12. Falcon LA, van den Bosch R, Ferris CA, Stromberg LK, Etzel LK et al. (1968) A comparison of season-long cotton-pest-control programs in California during 1966. J Econ Entomol 61: 633-642.
- View Article
- Google Scholar
13. Falcon LA, van den Bosch R, Gallagher J, Davidson A (1971) Investigation of the pest status of Lygus hesperus in cotton in central California. J Econ Entomol 64: 56-61.
- View Article
- Google Scholar
14. Sevacherian V, Stern VM (1972) Sequential sampling plans for Lygus bugs in California cotton fields. Environ Entomol 1: 704-710.
- View Article
- Google Scholar
15. Gutierrez AP, Leigh TF, Wang Y, Cave RD (1977) An analysis of cotton production in California: Lygus hesperus (Heteroptera: Miridae) injury – an evaluation. Canad Entomol 109: 1375-1386. doi:https://doi.org/10.4039/Ent1091375-10.
- View Article
- Google Scholar
16. Ellsworth PC (2000) Lygus control decision aids for Arizona cotton. Publ. No. AZ1170, University of Arizona. Available: http://cals.arizona.edu/pubs/crops/az1170/az11707b.pdf. Accessed 5 July 2013.
17. Ellsworth PC, Barkley V (2001) Cost-effective Lygus management in Arizona cotton. Publ. No. AZ1224, University of Arizona. Available: http://ag.arizona.edu/pubs/crops/az1224/az12247j.pdf. Accessed 5 July 2013.
18. Sadras VO (1995) Compensatory growth in cotton after loss of reproductive organs. Field Crops Res 40: 1-18. doi:https://doi.org/10.1016/0378-4290(94)00088-T.
- View Article
- Google Scholar
19. Wilson LJ, Sadras VO, Heimoana SC, Gibb D (2003) How to succeed by doing nothing: cotton compensation after simulated early season pest damage. Crop Sci 43: 2125-2134. doi:https://doi.org/10.2135/cropsci2003.2125.
- View Article
- Google Scholar
20. Lu Y, Wu K, Jiang Y, Xia B, Li P et al. (2010) Mirid bug outbreaks in multiple crops correlated with wide-scale adoption of Bt cotton in China. Science 328: 1151-1154. doi:https://doi.org/10.1126/science.1187881. PubMed: 20466880.
- View Article
- Google Scholar
21. Zeilinger AR, Olson DM, Andow DA (2011) Competition between stink bug and heliothine caterpillar pests on cotton at within-plant spatial scales. Entomol Exp Appl 141: 59-70. doi:https://doi.org/10.1111/j.1570-7458.2011.01165.x.
- View Article
- Google Scholar
22. Mangel M, Stefanou SE, Wilen JE (1985) Modeling Lygus hesperus injury to cotton yields. J Econ Entomol, 78: 1009-1014.
- View Article
- Google Scholar
23. Stewart SD, Sterling WL (1989) Causes and temporal patterns of cotton fruit abscission. J Econ Entomol 82: 954-959.
- View Article
- Google Scholar
24. Kerby TA, Weir BL, Keeley MP (1996) The uses of Pix. In: SJ HakeTA KerbyKD Hake. Cotton production manual. Oakland: University of California, Division of Agicultural and Natural Resources. pp. 294-304.
25. Rosenheim JA (2013) Costs of Lygus herbivory on cotton associated with farmer decision-making: an ecoinformatics approach. J Econ Entomol 106: 1286-1293. doi:https://doi.org/10.1603/EC12511. PubMed: 23865193.
- View Article
- Google Scholar
26. Dyer MI, Turner CL, Seastedt TR (1993) Herbivory and its consequences. Ecol Appl 3: 10-16. doi:https://doi.org/10.2307/1941781.
- View Article
- Google Scholar
27. Huhta AP, Hellström K, Rautio P, Tuomi J (2003) Grazing tolerance of Gentianella amarelle and other monocarpic herbs: why is tolerance highest at low damage levels? Plan Ecol 166: 49-61. doi:https://doi.org/10.1023/A:1023278502972.
- View Article
- Google Scholar
28. Gao Y, Wang D, Ba L, Bai Y, Liu B (2008) Interactions between herbivory and resource availability on grazing tolerance of Leymus chinensis. Environ Exp Bot 63: 113-122. doi:https://doi.org/10.1016/j.envexpbot.2007.10.030.
- View Article
- Google Scholar
29. Wood SN (2006) Generalized additive models: an introduction with R. Boca Raton. CRC Press. 392 pp.
30. Gross K, Rosenheim JA (2011) Quantifying secondary pest outbreaks in cotton and their monetary cost with causal-inference statistics. Ecol Appl 21: 2770-2780. doi:https://doi.org/10.1890/11-0118.1. PubMed: 22073658.
- View Article
- Google Scholar
31. Goodell PB (1998) Biology, ecology, and host plants of Lygus lineolaris and Lygus hesperus. Proc Beltwide Cotton Conf 2. pp. 949-951.
- View Article
- Google Scholar
32. Carriére Y, Ellsworth PC, Dutilleul P, Ellers-Kirk C, Barkley V, Antilla L (2006) A GIS-based approach for areawide pest management: the scales of Lygus hesperus movements to cotton from alfalfa, weeds, and cotton. Entomol Exp Appl 118: 203-210. doi:https://doi.org/10.1111/j.1570-7458.2006.00384.x.
- View Article
- Google Scholar
33. Carrière Y, Goodell PB, Ellers-Kirk C, Larocque G, Dutilleul P et al. (2012) Effects of local and landscape factors on population dynamics of a cotton pest. PLOS ONE 7: e39862. doi:https://doi.org/10.1371/journal.pone.0039862. PubMed: 22768147.
- View Article
- Google Scholar
34. Rosenheim JA, Steinmann K, Langellotto GA, Zink AG (2006) Estimating the impact of Lygus hesperus on cotton: the insect, plant, and human observer as sources of variability. Environ Entomol 35: 1141-1153. doi:https://doi.org/10.1603/0046-225X(2006)35[1141:ETIOLH]2.0.CO;2.
- View Article
- Google Scholar
35. Gutierrez AP, Falcon LA, Loew W, Leipzig PA, van den Bosch R (1975) An analysis of cotton production in California: a model for Acala cotton and the effects of defoliators on its yields. Environ Entomol 4: 125-136.
- View Article
- Google Scholar
36. Gutierrez AP, Wang Y, Regev U (1979) An optimization model for Lygus hesperus (Heteroptera: Miridae) damage in cotton: the economic threshold revisited. Canad Entomol 111: 41-54. doi:https://doi.org/10.4039/Ent11141-1.
- View Article
- Google Scholar
37. Kerby TA, Hake KD (1996) Monitoring cotton’s growth. In: SJ HakeTA KerbyKD Hake. Cotton production manual. Oakland: University of California, Division of Agricultural and Natural Resources. pp. 335-355.
38. Rochester WA, Zalucki MP, Ward A, Miles M, Murray DAH (2002) Testing insect movement theory: empirical analysis of pest data routinely collected from agricultural crops. Comput Electron Agric 35: 139-149. doi:https://doi.org/10.1016/S0168-1699(02)00015-7.
- View Article
- Google Scholar
39. de Valpine P, Scranton K, Ohmart CP (2010) Synchrony of population dynamics of two vineyard arthropods occurs at multiple spatial and temporal scales. Ecol Appl 20: 1926-1935. doi:https://doi.org/10.1890/09-0468.1. PubMed: 21049880.
- View Article
- Google Scholar
40. Cock J, Oberthür T, Isaacs C, Läderach PR, Palma A et al. (2011) Crop management based on field observations: Case studies in sugarcane and coffee. Agric Syst 104: 755-769. doi:https://doi.org/10.1016/j.agsy.2011.07.001.
- View Article
- Google Scholar
41. Jiménez D, Cock J, Jarvis A, Garcia J, Satizábal HF et al. (2011) Interpretation of commercial production information: A case study of lulo (Solanum quitoense), an under-researched Andean fruit. Agric Syst 104: 258-270. doi:https://doi.org/10.1016/j.agsy.2010.10.004.
- View Article
- Google Scholar
42. Steinmann KP, Zhang M, Grant JA (2011) Does use of pesticides known to harm natural enemies of spider mites (Acari: Tetranychidae) result in increased number of miticide applications? An examination of California walnut orchards. J Econ Entomol 104: 1496-1501. doi:https://doi.org/10.1603/EC11168. PubMed: 22066177.
- View Article
- Google Scholar
43. Parsa S, Ccanto R, Olivera E, Scurrah M, Alcázar et al . (2012) Explaining Andean potato weevils in relation to local and landscape features: a facilitated ecoinformatics approach. PLOS ONE 7(5): e36533. doi:https://doi.org/10.1371/journal.pone.0036533. PubMed: 22693551.
- View Article
- Google Scholar

[ref1] 1. Godfray HCJ, Beddington JR, Crute IR, Haddad L, Lawrence D et al. (2010) Food security: the challenge of feeding 9 billion people. Science 327: 812-818. doi:https://doi.org/10.1126/science.1185383. PubMed: 20110467.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Oerke EC (2006) Crop losses to pests. J Agric Sci 144: 31-43. doi:https://doi.org/10.1017/S0021859605005708.
View Article
Google Scholar

[5] View Article

[6] Google Scholar

[ref3] 3. Birch ANE, Begg GS, Squire GR (2011) How agro-ecological research helps to address food security issues under new IPM and pesticide reduction policies for global crop production systems. J Exp Bot 62: 3251-3261. doi:https://doi.org/10.1093/jxb/err064. PubMed: 21669880.
View Article
Google Scholar

[8] View Article

[9] Google Scholar

[ref4] 4. Rosenheim JA, Parsa S, Forbes AA, Krimmel WA, Law YH et al. (2011) Ecoinformatics for integrated pest management: expanding the applied insect ecologist’s tool-kit. J Econ Entomol 104: 331-342. doi:https://doi.org/10.1603/EC10380. PubMed: 21510177.
View Article
Google Scholar

[11] View Article

[12] Google Scholar

[ref5] 5. Stern VM, Smith RF, van den Bosch R, Hagen KS (1959) The integrated control concept. Hilgardia 29: 81-101.
View Article
Google Scholar

[14] View Article

[15] Google Scholar

[ref6] 6. Ragsdale DW, McCornack BP, Venette RC, Potter BD, MacRae IV et al. (2007) Economic threshold for soybean aphid (Hemiptera: Aphididae). J Econ Entomol 100: 1258-1267. doi:https://doi.org/10.1603/0022-0493(2007)100[1258:ETFSAH]2.0.CO;2. PubMed: 17849878.
View Article
Google Scholar

[17] View Article

[18] Google Scholar

[ref7] 7. Musser FR, Lorenz GM, Stewart SD, Bagwell RD, Leonard BR et al. (2009a) Tarnished plant bug (Hemiptera: Miridae) thresholds for cotton before bloom in the midsouth of the United States. J Econ Entomol 102: 2109-2115. doi:https://doi.org/10.1603/029.102.0614. PubMed: 20069839.
View Article
Google Scholar

[20] View Article

[21] Google Scholar

[ref8] 8. Musser FR, Catchot AL, Stewart SD, Bagwell RD, Lorenz GM et al. (2009b) Tarnished plant bug (Hemiptera: Miridae) thresholds and sampling comparisons for flowering cotton in the midsouthern United States. J Econ Entomol 102: 1827-1836. doi:https://doi.org/10.1603/029.102.0513. PubMed: 19886447.
View Article
Google Scholar

[23] View Article

[24] Google Scholar

[ref9] 9. Leigh TF, Kerby TA, Wynholds PF (1988) Cotton square damage by the plant bug, Lygus hesperus (Hemiptera: Heteroptera: Miridae), and abscission rates. J Econ Entomol 81: 1328-1337.
View Article
Google Scholar

[26] View Article

[27] Google Scholar

[ref10] 10. Forbes AA, Rosenheim JA (2011) Plant responses to insect herbivore damage are modulated by phosphorus nutrition. Entomol Exp Appl 139: 242-249. doi:https://doi.org/10.1111/j.1570-7458.2011.01125.x.
View Article
Google Scholar

[29] View Article

[30] Google Scholar

[ref11] 11. Cooper WR, Spurgeon DW (2012) Injury to cotton by adult Lygus hesperus (Hemiptera: Miridae) of different gender and reproductive states. Environ Entomol 41: 342-348. doi:https://doi.org/10.1603/EN11236. PubMed: 22507007.
View Article
Google Scholar

[32] View Article

[33] Google Scholar

[ref12] 12. Falcon LA, van den Bosch R, Ferris CA, Stromberg LK, Etzel LK et al. (1968) A comparison of season-long cotton-pest-control programs in California during 1966. J Econ Entomol 61: 633-642.
View Article
Google Scholar

[35] View Article

[36] Google Scholar

[ref13] 13. Falcon LA, van den Bosch R, Gallagher J, Davidson A (1971) Investigation of the pest status of Lygus hesperus in cotton in central California. J Econ Entomol 64: 56-61.
View Article
Google Scholar

[38] View Article

[39] Google Scholar

[ref14] 14. Sevacherian V, Stern VM (1972) Sequential sampling plans for Lygus bugs in California cotton fields. Environ Entomol 1: 704-710.
View Article
Google Scholar

[41] View Article

[42] Google Scholar

[ref15] 15. Gutierrez AP, Leigh TF, Wang Y, Cave RD (1977) An analysis of cotton production in California: Lygus hesperus (Heteroptera: Miridae) injury – an evaluation. Canad Entomol 109: 1375-1386. doi:https://doi.org/10.4039/Ent1091375-10.
View Article
Google Scholar

[44] View Article

[45] Google Scholar

[ref16] 16. Ellsworth PC (2000) Lygus control decision aids for Arizona cotton. Publ. No. AZ1170, University of Arizona. Available: http://cals.arizona.edu/pubs/crops/az1170/az11707b.pdf. Accessed 5 July 2013.

[ref17] 17. Ellsworth PC, Barkley V (2001) Cost-effective Lygus management in Arizona cotton. Publ. No. AZ1224, University of Arizona. Available: http://ag.arizona.edu/pubs/crops/az1224/az12247j.pdf. Accessed 5 July 2013.

[ref18] 18. Sadras VO (1995) Compensatory growth in cotton after loss of reproductive organs. Field Crops Res 40: 1-18. doi:https://doi.org/10.1016/0378-4290(94)00088-T.
View Article
Google Scholar

[49] View Article

[50] Google Scholar

[ref19] 19. Wilson LJ, Sadras VO, Heimoana SC, Gibb D (2003) How to succeed by doing nothing: cotton compensation after simulated early season pest damage. Crop Sci 43: 2125-2134. doi:https://doi.org/10.2135/cropsci2003.2125.
View Article
Google Scholar

[52] View Article

[53] Google Scholar

[ref20] 20. Lu Y, Wu K, Jiang Y, Xia B, Li P et al. (2010) Mirid bug outbreaks in multiple crops correlated with wide-scale adoption of Bt cotton in China. Science 328: 1151-1154. doi:https://doi.org/10.1126/science.1187881. PubMed: 20466880.
View Article
Google Scholar

[55] View Article

[56] Google Scholar

[ref21] 21. Zeilinger AR, Olson DM, Andow DA (2011) Competition between stink bug and heliothine caterpillar pests on cotton at within-plant spatial scales. Entomol Exp Appl 141: 59-70. doi:https://doi.org/10.1111/j.1570-7458.2011.01165.x.
View Article
Google Scholar

[58] View Article

[59] Google Scholar

[ref22] 22. Mangel M, Stefanou SE, Wilen JE (1985) Modeling Lygus hesperus injury to cotton yields. J Econ Entomol, 78: 1009-1014.
View Article
Google Scholar

[61] View Article

[62] Google Scholar

[ref23] 23. Stewart SD, Sterling WL (1989) Causes and temporal patterns of cotton fruit abscission. J Econ Entomol 82: 954-959.
View Article
Google Scholar

[64] View Article

[65] Google Scholar

[ref24] 24. Kerby TA, Weir BL, Keeley MP (1996) The uses of Pix. In: SJ HakeTA KerbyKD Hake. Cotton production manual. Oakland: University of California, Division of Agicultural and Natural Resources. pp. 294-304.

[ref25] 25. Rosenheim JA (2013) Costs of Lygus herbivory on cotton associated with farmer decision-making: an ecoinformatics approach. J Econ Entomol 106: 1286-1293. doi:https://doi.org/10.1603/EC12511. PubMed: 23865193.
View Article
Google Scholar

[68] View Article

[69] Google Scholar

[ref26] 26. Dyer MI, Turner CL, Seastedt TR (1993) Herbivory and its consequences. Ecol Appl 3: 10-16. doi:https://doi.org/10.2307/1941781.
View Article
Google Scholar

[71] View Article

[72] Google Scholar

[ref27] 27. Huhta AP, Hellström K, Rautio P, Tuomi J (2003) Grazing tolerance of Gentianella amarelle and other monocarpic herbs: why is tolerance highest at low damage levels? Plan Ecol 166: 49-61. doi:https://doi.org/10.1023/A:1023278502972.
View Article
Google Scholar

[74] View Article

[75] Google Scholar

[ref28] 28. Gao Y, Wang D, Ba L, Bai Y, Liu B (2008) Interactions between herbivory and resource availability on grazing tolerance of Leymus chinensis. Environ Exp Bot 63: 113-122. doi:https://doi.org/10.1016/j.envexpbot.2007.10.030.
View Article
Google Scholar

[77] View Article

[78] Google Scholar

[ref29] 29. Wood SN (2006) Generalized additive models: an introduction with R. Boca Raton. CRC Press. 392 pp.

[ref30] 30. Gross K, Rosenheim JA (2011) Quantifying secondary pest outbreaks in cotton and their monetary cost with causal-inference statistics. Ecol Appl 21: 2770-2780. doi:https://doi.org/10.1890/11-0118.1. PubMed: 22073658.
View Article
Google Scholar

[81] View Article

[82] Google Scholar

[ref31] 31. Goodell PB (1998) Biology, ecology, and host plants of Lygus lineolaris and Lygus hesperus. Proc Beltwide Cotton Conf 2. pp. 949-951.
View Article
Google Scholar

[84] View Article

[85] Google Scholar

[ref32] 32. Carriére Y, Ellsworth PC, Dutilleul P, Ellers-Kirk C, Barkley V, Antilla L (2006) A GIS-based approach for areawide pest management: the scales of Lygus hesperus movements to cotton from alfalfa, weeds, and cotton. Entomol Exp Appl 118: 203-210. doi:https://doi.org/10.1111/j.1570-7458.2006.00384.x.
View Article
Google Scholar

[87] View Article

[88] Google Scholar

[ref33] 33. Carrière Y, Goodell PB, Ellers-Kirk C, Larocque G, Dutilleul P et al. (2012) Effects of local and landscape factors on population dynamics of a cotton pest. PLOS ONE 7: e39862. doi:https://doi.org/10.1371/journal.pone.0039862. PubMed: 22768147.
View Article
Google Scholar

[90] View Article

[91] Google Scholar

[ref34] 34. Rosenheim JA, Steinmann K, Langellotto GA, Zink AG (2006) Estimating the impact of Lygus hesperus on cotton: the insect, plant, and human observer as sources of variability. Environ Entomol 35: 1141-1153. doi:https://doi.org/10.1603/0046-225X(2006)35[1141:ETIOLH]2.0.CO;2.
View Article
Google Scholar

[93] View Article

[94] Google Scholar

[ref35] 35. Gutierrez AP, Falcon LA, Loew W, Leipzig PA, van den Bosch R (1975) An analysis of cotton production in California: a model for Acala cotton and the effects of defoliators on its yields. Environ Entomol 4: 125-136.
View Article
Google Scholar

[96] View Article

[97] Google Scholar

[ref36] 36. Gutierrez AP, Wang Y, Regev U (1979) An optimization model for Lygus hesperus (Heteroptera: Miridae) damage in cotton: the economic threshold revisited. Canad Entomol 111: 41-54. doi:https://doi.org/10.4039/Ent11141-1.
View Article
Google Scholar

[99] View Article

[100] Google Scholar

[ref37] 37. Kerby TA, Hake KD (1996) Monitoring cotton’s growth. In: SJ HakeTA KerbyKD Hake. Cotton production manual. Oakland: University of California, Division of Agricultural and Natural Resources. pp. 335-355.

[ref38] 38. Rochester WA, Zalucki MP, Ward A, Miles M, Murray DAH (2002) Testing insect movement theory: empirical analysis of pest data routinely collected from agricultural crops. Comput Electron Agric 35: 139-149. doi:https://doi.org/10.1016/S0168-1699(02)00015-7.
View Article
Google Scholar

[103] View Article

[104] Google Scholar

[ref39] 39. de Valpine P, Scranton K, Ohmart CP (2010) Synchrony of population dynamics of two vineyard arthropods occurs at multiple spatial and temporal scales. Ecol Appl 20: 1926-1935. doi:https://doi.org/10.1890/09-0468.1. PubMed: 21049880.
View Article
Google Scholar

[106] View Article

[107] Google Scholar

[ref40] 40. Cock J, Oberthür T, Isaacs C, Läderach PR, Palma A et al. (2011) Crop management based on field observations: Case studies in sugarcane and coffee. Agric Syst 104: 755-769. doi:https://doi.org/10.1016/j.agsy.2011.07.001.
View Article
Google Scholar

[109] View Article

[110] Google Scholar

[ref41] 41. Jiménez D, Cock J, Jarvis A, Garcia J, Satizábal HF et al. (2011) Interpretation of commercial production information: A case study of lulo (Solanum quitoense), an under-researched Andean fruit. Agric Syst 104: 258-270. doi:https://doi.org/10.1016/j.agsy.2010.10.004.
View Article
Google Scholar

[112] View Article

[113] Google Scholar

[ref42] 42. Steinmann KP, Zhang M, Grant JA (2011) Does use of pesticides known to harm natural enemies of spider mites (Acari: Tetranychidae) result in increased number of miticide applications? An examination of California walnut orchards. J Econ Entomol 104: 1496-1501. doi:https://doi.org/10.1603/EC11168. PubMed: 22066177.
View Article
Google Scholar

[115] View Article

[116] Google Scholar

[ref43] 43. Parsa S, Ccanto R, Olivera E, Scurrah M, Alcázar et al . (2012) Explaining Andean potato weevils in relation to local and landscape features: a facilitated ecoinformatics approach. PLOS ONE 7(5): e36533. doi:https://doi.org/10.1371/journal.pone.0036533. PubMed: 22693551.
View Article
Google Scholar

[118] View Article

[119] Google Scholar

Figures

Abstract

Introduction

Materials and Methods

Statistical analysis

Results

A positive control for data quality

Lygus hesperus densities and Gossypium yield

Collinearity of June and July L. hesperus densities

Searching for possible confounding variables

Discussion

Yield gaps: evidence from modeling and experimentation

Unnecessary use of insecticides

Plant compensation by extending the growing season

Conclusions

Supporting Information

Acknowledgments

Author Contributions

References