Two large-scale experiments were conducted in two commercial citrus groves in the southwest Florida citrus growing area (Hendry County, Florida, USA) from summer 2010 through spring 2014. The two experimental sites were located in large citrus groves and consisted of distinct blocks of orange trees individually managed in which owners and managers agreed to cede all pest management decisions to project managers. Site 1: A 10.3 ha block in a grove belonging to Bob Paul Inc. near LaBelle (26°41’04”N, 81°26’20”W), planted December 2001 with early season sweet oranges Citrus sinensis (L.) Obseck ‘Earlygold’, bud-grafted to ‘Carrizo’ citrange rootstock at a density of 231 trees ha^-1. Site 2: A 5.4 ha block located in a large orchard belonging to Moreno Farms near Felda (26° 34' 0.6" N, 81° 26' 20.1"W) and planted in 1999 with late season sweet oranges Citrus sinensis (L.) Obseck ‘Valencia’ on ‘Swingle’ citrumelo rootstock at a density of 336 tree ha^-1. Management followed conventional cultural practices [30] except for insecticidal control. In addition, a HLB foliar nutritional remediation program was applied throughout each block by the growers during major tree flushes (Table 1).

The two study sites were divided into 16 plots (experimental units), each containing approximately 144 and 120 trees at sites 1 and 2 respectively. A randomized complete block design was used, each with four replications and 4 treatments: (1) ‘calendar’—monthly sprays of insecticides to control ACP, (2) ‘0.2 thsld’—insecticide applications based on a nominal threshold of 0.2 ACP adults per stem-tap sample plus two “dormant” (winter) sprays, (3) ‘0.7 thsld’—insecticide applications based on nominal threshold of 0.7 ACP adults per stem-tap sample plus one dormant spray, and (4) ‘no insecticide’—a control treatment where no insecticide applications to control ACP. The ‘calendar’ treatment was aimed to keep ACP densities as low as possible by following an intensive spray schedule using a rotation of insecticides. At the other extreme was the ‘no insecticide’ control treatment where ACP populations were allowed to vary without insecticide intervention. The two nominal threshold treatments were set to create intermediate ACP densities and control costs.

All Insecticide treatments were made in designated plots with an Air-O-Fan airblast sprayer equipped with Albuz^® ATR hollow cone nozzles providing an 80° spray pattern with five nozzles, two of 2.5 and three of 4 L/min, operating at 300 PSI and 3 km/h delivering a total of 887 L/ha. Monthly applications against ACP were initiated in 2010 in plots designated for ‘calendar’ treatment of both study sites (Tables 2 and 3). Broad-spectrum products (organophosphates, carbamates and pyrethroids) were generally restricted to the winter and the end of the summer whereas more selective insecticides were chosen preferentially during the growing season to rotate modes of action, control other pest present and reduce impacts on beneficial arthropod fauna [17–19, 21, 31]. Ten modes of action were rotated to avoid inducing resistance in D. citri. Thus, insecticide choices during the growing season were made on the basis of efficacy, timeliness, and resistance management rather than product cost. Occasional sprays required for control of other pests or diseases were chosen for minimal impact on ACP and made over the entire study site to avoid confounding treatment effects.

Sprays for the two additional treatments (‘0.2 thsld’ and ‘0.7 thsld’) were triggered during the growing season if and when tap samples averaged over all 4 replicates reached or exceeded nominal thresholds. At that time, all 4 replicates of the treatment were sprayed using the same product chosen for the ‘calendar’ treatment that month. Dormant season insecticide applications in these two treatments were made without consideration of thresholds in conformance with synchronized dormant sprays organized in the local “citrus health management area” [CHMA www.flchma.org]. The ‘0.7 thsld’ treatment received its dormant spray in January whereas ‘0.2 thsld’ treatment received two dormant sprays, one in December and one in January. Each site was sprayed at least once per season with copper-based products to control citrus canker, Xanthomonas axonopodis pv. citri. Site 1 was sprayed by the grower in August 2011 and October 2012 with sulfur (Microthiol Disperss^®) to control citrus rustmite, Phyllocoptruta oleivora (Ashmead) (Acari: Eriophyidae). Clorpyrifos (Lorsban^® 4EC) was applied inadvertently by the grower over all of Site 2 for general pest control in March 2012.

Density of ACP adults was monitored every other week from 7 July 2010 through 6 November 2013 at Site 1 and from 12 August 2010 to 24 February 2014 at Site 2 by stem-tap on two sides of 10 randomly selected trees in the vicinity of two previously selected stops per plot (320 trees sampled per grove and date) [23]. A 22 × 28 cm plastic laminated white paper sheet was held horizontally about 30 cm underneath a randomly chosen branch, which was then struck sharply three times with a length of PVC pipe. ACP adults falling on the sheet were quickly counted and the two numbers averaged to obtain the sample unit: “ACP adults per stem-tap and tree” [23]. Biweekly mean numbers of ACP adults per stem-tap were used to determine whether insecticide applications were necessary in plots designated for the ‘0.2 thsld’ and ‘0.7 thsld’ treatments. ACP adult cumulative numbers per stem-tap and tree (κ) were summed over the season through each harvest date in fall for ‘Earlygold’ (Site 1) and spring for ‘Valencia’ (Site 2).

The proportion of trees testing positive for Candidatus Liberibacter asiaticus in each treatment was evaluated twice per year, in midsummer and at the end of winter, beginning in summer 2010 before treatments were initiated, through summer 2013. Samples consisted of 20 randomly collected leaves per tree, five at each cardinal point, from 20 randomly sampled trees per plot. Samples were taken to the laboratory and DNA extracted from a 50 mg dry weight subsample of petiole tissue using the Promega Wizard ^® 96 DNA Plant isolation kit (Promega, USA). Real-time qPCR was conducted with an ABI 7500 Fast Real-Time PCR System (Applied Biosystems, Foster City, CA) in a 20 μl volume using HLBas/HLBr and HLBp Candidatus Liberibacter asiaticus primers [32]. The standard amplification protocol was initial denaturation at 95°C followed by 40 cycles of reactions (95°C for 3 s, 60°C for 30 s). Data was analyzed using Applied Biosystems 7500 system SDS software. Samples were considered positive for Ct values less or equal to 36 [33].

All marketable fruit was harvested by plot in December 7–13 in 2010 (2010–11 season), November 16–20 in 2011 (2011–12 season), November 14–23 in 2012 (2012–13 season) and November 18–21 in 2013 (2013–14 season) in Site 1, and January 28 to February 3 in 2011 (2010–11 season), January 23 to February 3 in 2012 (2011–12 season), in February 24–26 in 2013 (2012–13 season) and March 6–12 2014 (2013–14 season) in Site 2. The mass of fruit from each plot (kg of fruit per ha) was estimated by counting full and partially full tubs during the harvest. Tubs were 784.1 dm³ in volume and designed to contain a volume of ten standard boxes of oranges (40.8 kg per box) [34] when totally full. Partially full tubs were measured from the top of the container to the level of harvested fruit and the corresponding fruit volume and weight was calculated.

Juice quality, measured as ‘kg of juice per kg of fruit’, ‘kg of soluble solids per kg of fruit’, acidity, ‘total degrees brix’ and ‘brix/acid’ ratio, was evaluated from a 2 bushel composite random fruit sample collected in each plot one week prior to initiating harvest. Samples were brought to the University of Florida citrus quality laboratory in Lake Alfred, FL where juice was extracted and de-aerated under vacuum for 2–3 minutes. Soluble solids content was measured by hydrometer and titratable acidity as citric acid, pH endpoint 8.2. Since production destined for the juice industry and processed oranges prices are valued by the amount of soluble solids, yields (kg of solids per ha) were calculated by multiplying the harvested kg of fruit per ha from each plot by estimated solids (kg of soluble solids per kg of fruit) obtained from the corresponding treatment through juice analyses. When no significant treatment effect was found, the average value for the entire grove was used for calculations.

The economic viability of each tested treatment was studied using yield data from all harvests. Incomes for processed oranges ($/ha) were estimated for the different treatments in each year by comparing changes in revenues under a range of fruit prices representing those obtained for early and late-season juice oranges in Florida during the last three seasons [35].

Production costs were divided into the following grove operations [36]: pest management, foliar nutrition program, canker management, fruit picking, hauling and Florida Department of Citrus assessment costs, ACP monitoring, weed management, ground fertilizer, pruning practices, irrigation and tree replacement costs (Table 4). Property and district taxes, management fees charged by professional caretakers, interest charges on operating capital and interest on investment capital were also included as indirect production costs. Foliar chemical application costs for pest and disease management were separated into material and application costs (Tables 2 and 3). Product prices were provided by chemical and fertilizer vendors. Application costs ($59 per ha) were those estimated by Roka et al. [36] for 935 L per ha (100 gallon per acre) air-blast chemical applications in Central and Southwest Florida. Similarly, we used the same estimate as these authors for all harvesting costs (Pick and roadside: $0.063 per kg of harvested fruit; haul: $0.018 per kg of fruit; FDOC assessment: $0.006 per kg of fruit). ACP scouting costs were estimated taking into account the time spent on monitoring each plot and were only applied to those treatments where the decision to spray was based on nominal thresholds (0.2 and 0.7 thsld). For the remaining operations, we used costs estimated by Roka et al. [36] for orange juice production in southwest Florida.

Based on total management costs and yields, a break-even price ($/kg of solids) to recover production costs was calculated for each site, treatment and season and for the last three seasons together. Profitability ($/ha) was also calculated by subtracting total management costs from income obtained at the different price scenarios evaluated.

Differences in ACP adult cumulative numbers among treatments through harvest for each year and site were assessed using repeated measures analysis based on generalized mixed models. Data was assumed to be Poisson distributed; ‘block’ (replicate) was considered as a random factor, ‘treatment’ as fixed factor and ‘time’ as the repeated fixed effect. Several covariance structures were tested and an autoregressive structure was selected based on Akaikei and Bayesian information criteria (AIC and BIC respectively). Treatment effects on the proportion of trees testing positive for HLB by qPCR for each sample date were evaluated for Site 2 by generalized mixed model analysis where data was assumed to be binomially distributed, ‘block’ was considered as a random factor and ‘treatment’ as a fixed factor. Treatment effects on juice quality parameters were assessed using general mixed model analyses where ‘block’ was considered as a random factor and ‘treatment’ as a fixed factor. Changes through time in yields and differences in this parameter among treatments were analyzed for each Site through repeated measures analysis using generalized mixed models. Data from Site 1 was found to be normally distributed whereas data from Site 2 were slightly overdispersed and a Poisson distribution assumption for data resulted in better fit based in the scaled Pearson statistic and Akaikei and Bayesian information criteria. ‘Block’ was considered as a random factor, ‘treatment’ as fixed factor and ‘time’ as the repeated fixed effect. Autoregressive and heterogeneous compound symmetric covariance structures were selected for Sites 1 and 2 respectively based on AIC and BIC. Post-hoc t-test (LSD) comparisons were made in case of any significant effect (P < 0.05) for all analyses.

A model was developed for each site and for pooled data from both sites to predict yield losses caused by the pest in 10–15 year-old blocks of orange trees under moderate to high HLB incidence. The model used 2013–14 harvest results to forecast a target number of psyllids per season that would balance costs of a given insecticidal ACP management program. Economic injury level was calculated as the vector density at which additional costs for ACP management were equal to the estimated losses caused by this vector-disease system [28] using Eq (1). C=P⋅Ym⋅ρ (1) Where C in $/ha represents ACP management costs, P is the harvest price in $/kg of solids and ρ is the effect on the ACP population on maximum yield Y_m when the pest is not present or at its lowest practical level. In our study, Y_m was estimated based on the average yield from the last harvest (2013–14) in the ‘calendar’ plots expressed as kg of solids per ha, where ACP numbers were as close to 0 as could reasonably be expected. Proportional yield losses for any of the other 12 plots (i) of each grove (ρ) were calculated with respect to the average yield for the ‘calendar’ treatment as: ρ=Ym−YiYm (2)

The variable ρ represented yield loss without indicating any particular cause. Non-linear least-squares regression analysis was used to explain the relationship between yield losses and ACP cumulative number per tree and season (κ) for each plot from the beginning of the experiments until the 2013–14 harvest. Data was fitted to a rectangular hyperbolic Eq (3) [28, 37–39] by using the Newton-Raphson iterative estimation procedure that included two parameters: ‘I’ and ‘A’: ρ= I⋅κ1+I⋅κA⋅100−1 (3) Where ‘I’ was defined as the horizontal asymptote of the function, which in our study represented the maximum yield loss (%) that the pest could induce after four seasons. ‘A’, was the slope of the curve at the origin which represented the rate of yield loss at low pest density. Results from two plots in each site were considered outliers and therefore removed from the equation.

Once ρ was expressed as a function of vector densities, Eq (1) was written as: C=P⋅Ym⋅(I⋅k1+I⋅kA)⋅100−1 (4)

Given known values for C, P and Y_m, the vector density, expressed as ACP adult cumulative number per tree and season (κ) that balanced the two sides of Eq (4) would define the EIL for these conditions.

An insecticide vector control approach was adopted, based on yield losses and ACP densities relationships, and taking into account the variable nature of management costs and juice market prices. The decision to spray during the growing season in mature citrus groves under high HLB incidence would be made according to ACP data collected during that season through stem-tap sampling. The program would commence with two dormant-season spray applications during winter, irrespective to ACP densities [21, 40]. Meanwhile, a running estimate would be maintained charting the number of ACP adults over the season and expressed as ACP cumulative number per stem. It would be time to spray again when the sum of costs of the two dormant-season sprays, a first insecticide application during the growing season and scouting equaled estimated losses based on ACP numbers. A new threshold would then be calculated to balance the costs listed above plus the additional application. The procedure would be repeated until the end of the growing season.

Insecticide management significantly reduced ACP adult abundance at both sites (treatment effect: F = 17.88; df = 3, 69; P < 0.0001 for Site 1, and F = 12.71; df = 3, 27.9; P < 0.0001 for Site 2). Lowest ACP cumulative values were seen at both sites throughout the study in plots receiving calendar insecticide applications against the psyllid compared to the untreated control or the ‘0.7 thsld’ treatment (Fig 1). A significant interaction effect between treatment and time at Site 2 (F = 3.57; df = 9, 102.8; P = 0.0007) reflected the fact higher ACP numbers were seen early on in ‘0.7 thsld’ treatment plots compared to ‘no insecticide’ treatment plots, although the situation reversed toward the end of the study. This effect was nevertheless not seen at Site 1 (F = 1.17; df = 9, 69; P = 0.3282).

At Site 1, 1,117 of the 1,120 samples analyzed by qPCR throughout the study were positive for C. L. asiaticus indicating that practically 100% of the trees on this study site were HLB infected from the onset of the experiment. In Site 2, HLB incidence ranged between 10 and 50% of trees testing positive at the beginning of the experiment with differences between treatments (F = 9.50; df = 3, 25; P = 0.0004) associated with uncontrolled variability in the grove such as edge effects. Incidence increased with time with no significant treatment effects until the last sampling date (March 2011: F = 1.97; df = 3, 25; P = 0.1436; September 2011: F = 2.71; df = 3, 25; P = 0.0668; March 2012: F = 1.03; df = 3, 25; P = 0.3968; September 2012: F = 0.35; df = 3, 25; P = 0.7918; March 2013: F = 0.65; df = 3, 25; P = 0.5899; September 2013: F = 3.49; df = 3, 25; P = 0.0304, Fig 2). At the end of the experiment, all sampled trees in the ‘no insecticide’ treatment tested positive for HLB compared to 88.5 ± 6.4% in trees subjected to the ‘0.7 thsld’, 80.0 ± 6.5% to the ‘0.2 thsld’ and 67.5 ± 10.0% for the calendar treatment, the latter being significantly less than either the ‘no insecticide’ or ‘0.7 thsld’ treatments.

No treatment effects were observed on juice per kg fruit, solids per fruit or acidity for any of the harvests in Site 1 (Table 5). However, the brix:acid ratio was lower in fruit harvested in 2011 and brix was higher in 2013 in fruit from trees receiving calendar sprays compared to untreated or ‘0.2 thsld’ treatments and untreated and ‘0.7 thsld’ respectively, although no such effects were observed at any other harvest (Table 5). At Site 2, significantly less juice per kg fruit was obtained from ‘calendar’ sprayed trees compared to the remaining treatments harvested in 2012. Similarly, soluble solids were significantly less in fruit from ‘calendar’ plots compared to the rest in 2011 and 2012 and for acid in 2012. No significant treatment effects were found at the last (2014) harvest nor for total degrees brix and brix/acid ratio parameters from any of the four harvests.

Significant effects on yield (kg solids per ha) were found in Site 1 (‘Earlygold’) for all but the first (2010) harvest and over all 4 harvests (F = 4.27; df = 3, 12.85; P = 0.0267, Table 6). ‘Calendar’ treatments resulted in greatest yields at all of the 3 last harvests, although not significantly different from the control or ‘0.7 thsld’ in 2012–13. Time effects were significant (F = 11.68; df = 3, 36.36; P < 0.0001) but some varied among treatments (treatment x time interaction: F = 2.00; df = 9, 36.70; P = 0.067). Yields from trees sprayed by the ‘calendar’ or 0.2 threshold increased over the 4 year study (‘calendar’: t-value = -4.44; df = 44.32; P < 0.0001; ‘0.2 thsld’: t-value = -3.63; df = 44.32; P = 0.0007), but marginally from unsprayed trees (t-value = -1.92; df = 44.32; P = 0.061). No yield increases were observed in trees sprayed on a 0.7 threshold (t-value = -1.32; df = 44.32; P = 0.19).

Treatment effects in Site 2 (‘Valencia’) were significant for the last two harvests. Again, highest yields were seen from the ‘calendar’ treatment although not significantly different from ‘02 thsld’ (Table 6). Although no treatment effects were observed over all 4 harvests (F = 0.41; df = 3, 45; P = 0.7472), the treatment x time interaction was significant (F = 2.50; df = 9, 45; P = 0.0205) with a general trend toward increasing yields from both high input treatments for the 2012 and 2013 harvests (‘calendar’: t-value = -2.39; df = 45.00; P = 0.0211, ‘0.2 thsld’: t-value = -3.84; df = 45.00; P = 0.0004) that was not observed with the two low input treatments (‘0.7 thsld’: t-value = -0.78; df = 45.00; P = 0.4394, ‘no insecticide’: t-value = -0.69; df = 45.00; P = 0.4923). However, the trend varied at the last (2014) harvest when a sharp increase in ACP densities that season coincided with reduced yields from all treatments with respect to the previous season (‘calendar’: t-value 2.78; df = 45.00; P = 0.0079, ‘0.2 thsld’: t-value = 2.80; df = 45.00; P = 0.0075, ‘0.7 thsld’: t-value = 2.67; df = 45.00; P = 0.0106, ‘no insecticide’: t-value = 6.70; df = 45.00; P < 0.001). Yield reduction was most notable in ‘no insecticide’ plots (Table 6).

‘Calendar’ ACP insecticide management increased grove caretaking costs over the untreated by approximately $1,500 per ha and season. Treatment costs based on nominal thresholds were lower than calendar sprays, but varied from year to year depending on ACP populations (Table 4). Delivered-in break-even prices covering all production costs were highest ($5.14 per kg solids) and lowest ($3.89 per kg solids) in untreated plots, for the 2011–2012 and 2013–2014 seasons respectively. This contrasted with results from Site 2 where extremes of $5.22 and $3.66 the 2010–2011 and 2013–2014 harvests resulted instead from the 0.2 threshold treatment (Table 7). When costs and revenues from the last three harvests were summed, break-even prices were lowest with the ‘0.7 thsld’ ($4.42) and highest with the ‘0.2 thsld’ ($4.56) treatments in Site 1. Opposite results were obtained at Site 2: highest with the ‘0.7 thsld’ ($4.13) and lowest with the ‘0.2 thsld’ ($3.79).

At the low ($3.79/kg solids) price, none of the 4 management strategies would have been profitable for any season in either site except the ‘0.2 thsld’ treatment at Site 2 for the 2013–2014 season and for the sum of the last three seasons. In contrast, all treatments were profitable in all seasons at the high $5.24 per kg solids scenario (Table 7). In general, strategies incurring higher ACP management costs became relatively more profitable as juice prices increased, although profitability varied among groves, treatments and seasons. For example, the ‘0.2 thsld’ strategy resulted most profitable at Site 2 for all but the first season and over all dates but never at Site 1.

Average yields in plots receiving monthly calendar insecticide applications (Y_m) were 2,014.5 ± 51.6, 1,962.0 ± 245.4 and 1,988.3 ± 75.2 kg of soluble solids per ha in Site 1, Site 2, and pooled over the two groves respectively. Using the ‘calendar’ plots (Y_m) as a baseline, the rectangular hyperbolic model indicated a significant relationship between yield losses (ρ) in each plot on the one hand, and ACP adult cumulative numbers per stem-tap and tree (κ) on the other, for either site alone and pooled (Site 1: F = 25.00; df = 2, 14; P < 0.0001; Pseudo-R² = 0.46; Site 2: F = 24.48; df = 2, 14; P < 0.0001; Pseudo-R² = 0.68; two sites pooled: F = 38.10; df = 2, 30; P < 0.0001; Pseudo-R² = 0.52) (Fig 3). Estimated slopes at the origin were I = 3.39 ± 1.85 for Site 1, I = 1.03 ± 0.49 for Site 2 and I = 0.97 ± 0.38 for the two groves pooled. The estimated horizontal asymptotes that define maximum yield losses (%) were A = 21.80 ± 4.68 for Site 1, A = 59.06 ± 21.61 for Site 2 and A = 50.91 ± 18.69 for the two groves pooled. Substituting these values for ‘Y_m’, ‘I’ and ‘A’ values in Eq (4) provided the following equations for Site 1 Eq (5.1), Site 2 Eq (5.2) and the two groves pooled Eq (5.3): C=P⋅2014.5⋅(3.39⋅k1+3.39⋅k21.8)⋅100−1 (5.1) C=P⋅1962⋅(1.03⋅k1+1.03⋅k59.06)⋅100−1 (5.2) C=P⋅1988.3⋅(0.97⋅k1+0.97⋅k50.91)⋅100−1 (5.3)

Given known ACP management costs (C) and juice prices (P), the EIL could thus be defined by the cumulative number of ACP adults per stem-tap, tree and season (κ) that balanced the two sides of these equations [S1 File].

The cumulative number of adults per stem-tap, tree and season over which it would be more cost-effective to spray than not to spray is calculated from Eq (5) considering an orange juice price of $5.25 per kg of solids ($2.38 per pound-solids), approximating spraying costs at $75/ha and $130/ha for dormant and growing season respectively, and allocating $141 per ha and season for ACP monitoring (Table 8).